SlideShare a Scribd company logo
1 of 8
Download to read offline
Hybrid Periodical Flooding in Unstructured Peer-to-Peer Networks*
*
This work was partially supported by Michigan State University IRGP Grant 41114 and by Hong Kong RGC Grant HKUST6161/03E.
Zhenyun Zhuang1
, Yunhao Liu1
, Li Xiao1
and Lionel M. Ni2
1
Department of Computer Science and Engineering, Michigan State University, U.S.A.
2
Department of Computer Science, Hong Kong University of Science and Technology, Hong Kong
{zhuangz1, liuyunha, lxiao}@cse.msu.edu, ni@cs.ust.hk
Abstract
Blind flooding is a popular search mechanism used
in current commercial P2P systems because of its sim-
plicity. However, blind flooding among peers or super-
peers causes large volume of unnecessary traffic al-
though the response time is short. Some improved sta-
tistics-based search mechanisms can reduce the traffic
volume but also significantly shrink the query coverage
range. In some search mechanisms, not all peers may
be reachable creating the so-called partial coverage
problem. Aiming at alleviating the partial coverage
problem and reducing the unnecessary traffic, we pro-
pose an efficient and adaptive search mechanism, Hy-
brid Periodical Flooding (HPF). HPF retains the ad-
vantages of statistics-based search mechanisms, allevi-
ates the partial coverage problem, and provides the
flexibility to adaptively adjust different parameters to
meet different performance requirements. The effective-
ness of HPF is demonstrated through simulation studies.
1 Introduction
In an unstructured P2P system, such as Gnutella [7]
and KaZaA [8], file placement is random, which has no
correlation with the network topology [17]. Unstruc-
tured P2P systems are most commonly used in today's
Internet. In an unstructured P2P system, when a source
peer needs to query an object, it sends a query to its
neighbors. If a peer receiving the query cannot provide
the requested object, it may relay the query to its own
neighbors. If the peer receiving the query can provide
the requested object, a response message will be sent
back to the source peer along the inverse of the query
path. The most popular query operation in use, such as
Gnutella and KaZaA (among supernodes), is to blindly
“flood" a query to the network. A query is broadcast
and rebroadcast until a certain criterion is satisfied. This
mechanism ensures that the query will be “flooded” to
as many peers as possible within a short period of time
in a P2P overlay network. However, flooding also
causes a lot of network traffic and most of which is un-
necessary. Study in [13] shows that P2P traffic contrib-
utes the largest portion of the Internet traffic based on
their measurements on three popular P2P systems,
FastTrack (including KaZaA and Grokster) [5],
Gnutella, and DirectConnect. The inefficient blind
flooding search technique causes the unstructured P2P
systems being far from scalable [11].
To avoid the large volume of unnecessary traffic in-
curred by flooding-based search, many efforts have
been made to improve search algorithms for unstruc-
tured P2P systems. One typical approach is statistics-
based, in which instead of flooding to all immediate
overlay neighbors, a peer selects only a subset of its
neighbors to query based on some statistics information
of some metrics and heuristic algorithms. When han-
dling a query message (either relayed from its neighbor
or originated from itself) in a statistics-based search al-
gorithm, the peer determines the subset of its logical
neighbors to relay the query message. Statistics-based
search mechanisms may significantly reduce the traffic
volume but may also reduce the query coverage range
so that a query may traverse a longer path to be satisfied
or cannot be satisfied. In some search mechanisms, not
all peers may be reachable creating the so-called partial
coverage problem. Our objective is trying to alleviate
the partial coverage problem and reduce unnecessary
traffic.
In this paper, Section 2 will give an overview and
classification of known search mechanisms. The con-
cept of our proposed periodical flooding method will be
introduced in Section 3. Based on periodical flooding
and weighted metrics in selecting relay neighbors, the
hybrid periodical flooding (HPF) method is detailed in
Section 3. The proposed HPF can improve the effi-
ciency of blind flooding by retaining the advantages of
statistics-based search mechanisms and by alleviating
the partial coverage problem. Section 4 describes our
simulation method and the performance metrics. Per-
formance evaluation of our proposed HPF method
against other search methods is described in Section 5.
Section 6 concludes the paper.
2 Search Mechanisms
In unstructured P2P systems, the placement of ob-
jects is loosely controlled and each peer has no hint
where the intended objects are stored. Without having
the global knowledge of the dynamic overlay network
and the locations of target peers, a source peer has to
send a query message to explore as many peers as pos-
sible in the overlay network. A well-designed search
mechanism should seek to optimize both efficiency and
Quality of Service (QoS). Efficiency focuses on better
utilizing resources, such as bandwidth and processing
power, while QoS focuses on user-perceived qualities,
such as number of returned results and response time.
In unstructured P2P systems, the QoS of a search
mechanism generally depends on the number of peers
being explored (queried), response time, and traffic
overhead. If more peers can be queried by a certain
query, it is more likely that the requested object can be
found. In order to avoid having query messages flowing
around the network forever, each query message has a
TTL (time-to-live: the number of times a query will be
forwarded) field. A TTL value is set to limit the search
depth of a query. Each time a peer receives a query, the
TTL value is decremented by one. The peer will stop
relaying the query if TTL becomes zero. A query mes-
sage will also be dropped if the query message has vis-
ited the peer before. Note that the query messages are
application-level messages in an overlay network.
In statistics-based search mechanisms, a peer selects
a subset of its neighbors to relay the query based on
some statistics information of some metrics and heuris-
tic algorithms. Based on the number of selected logical
query neighbors and the criteria in selecting logical
query neighbors, the statistics-based search algorithms
in unstructured P2P systems can be roughly classified
into two types: uniformed selection of relay neighbors
and weighted selection of relay neighbors.
2.1 Uniformed Selection of Relay Neighbors
In this approach, all logical neighbors are equally
treated when selected to relay the query message.
Blind flooding. Blind flooding mechanism relays the
query message to all its logical neighbors, except the
incoming peer. This mechanism is also referred as
breadth-first search (BFS) and is used among peers in
Gnutella or among supernodes in KaZaA. For each
query, each node records the neighbors which relay the
query to it. Thereby on each link, at most two query
messages can be sent across it. For an overlay network
with m peers and average n neighbors per peer, the total
traffic caused by a query is mn if the value of TTL is no
less than the diameter of the overlay network. Note that
in a typical P2P system, the value of m (more than mil-
lions) is much greater than n (less than tens) [13]. In
this approach, the source peer can reach its target peer
(object) through a shortest path. However, the overhead
of blind flooding is very large since flooding generates
large amount of unnecessary traffic, wasting bandwidth
and processing resource. The simplicity of blind flood-
ing makes it very popular in practice.
Depth-first search (DFS). Instead of sending que-
ries to all the neighbors, a peer just randomly selects a
single neighbor to relay the query message when the
TTL value is not zero and waits for the response. This
search mechanism is referred to as depth-first search
(DFS) and is used in Freenet [6]. DFS can terminate
timely when the required object has been found, thus
avoiding sending out too many unnecessary queries. In
DFS, the value of TTL should be set sufficiently large
to increase the probability of locating the object. The
maximum number of peers that a query message will
visit is TTL. Thus, setting a proper TTL value is a key
issue to determine the search quality. The response
time could be unbearably large due to the nature of its
sequential search process. Because of the random selec-
tion of relay neighbors, it is possible that an object can
hardly be found.
K-walker. In k-walker query algorithm proposed in
[10], a query is sent to k different walkers (relay
neighbors) from the source peer. For a peer in each
walker, it just randomly selects one neighbor to relay
the query. For each walker, the query processing is
done sequentially. For k walkers with up to TTL steps,
each query can reach up to k�TTL peers in the P2P
network. We can view k-walker search mechanism as a
multiple of DFS. It has been shown that k-walker
mechanism creates less traffic than that of BFS and
provides shorter response time than that of DFS. How-
ever, k-walker suffers limited query coverage range due
to the randomness nature in selecting query neighbors.
2.2 Weighted Selection of Relay Neighbors
Instead of randomly selecting relay neighbors, some
mechanisms have been proposed to select relay
neighbors more objectively so that neighbors who are
most likely to return the requested results are selected.
Some statistics information is collected based on some
metrics when selecting relay neighbors. Possible met-
rics include delay of the link to the corresponding
neighbor, the processing time of the neighbor, the com-
puting power, the cost (if possible), the amount of shar-
ing data, and the number of neighbors, etc.
Directed BFS (DBFS). Each peer maintains statistic
information based on some metrics, such as the number
of results received from neighbors from previous que-
ries or the latency of the connection with that neighbor.
A peer selects a subset of the neighbors to send its
query based on some heuristics, such as selecting the
neighbors that have returned the largest number of re-
sults from previous queries or selecting the neighbors
that have the smaller latency.
Routing indices (RI). The concept of routing indices
(RI) was proposed in [3]. Each peer keeps a local RI
that is a detailed summary of indices, such as the num-
ber of files on different topics of interests along each
path. When a peer receives a query, it forwards the
query to the neighbor that has the largest number of
files under a particular topic, rather than selecting relay
neighbors at random or flooding to all neighbors.
Some weighted-selection search mechanisms have
demonstrated performance improvement compared with
uniformed-selection search mechanisms. However,
weighted-selection search mechanisms have the partial
coverage problem to be illustrated in Section 2.4.
2.3 Other Approaches
In addition to the aforementioned search policies,
there are other techniques that may be used to improve
search performance. For example, a peer can cache
query responses in hoping that subsequent queries can
be satisfied quickly by the cached indices or responses
[14, 16, 17]. Peers can also be clustered based on dif-
ferent criteria, such as similar interests [14], location in-
formation [9], and associative rules [4]. Our proposed
statistics-based technique can be used to complement
these techniques.
2.4 Partial Coverage Problem
Statistics-based search algorithms indeed can reduce
network traffic. For example, compared with blind
flooding, DBFS can reduce the aggregate processing
and bandwidth cost to about 28% and 38%, respectively
with 40% increase in the response time [17]. However,
our study will show that statistics-based search mecha-
nisms may leave a large percentage of the peers un-
reachable no matter how large the TTL value is set. We
call this phenomena partial coverage problem. This
problem is illustrated in Fig.1(a). The number by an
edge is the latency between two logical nodes and the
number in each node is the number of shared files on
that peer. Suppose the size of selected neighbor subset
is one and the metric used to select the neighbor is
based on the number of shared files. We consider the
scenario when the query source is A who has four
neighbors (B, C, D, E). It will only send its query to C
since C has the largest number of shared files (170).
Similarly, C selects D who has the largest number of
shared files in all C’s neighbors (B, D, F, G) to relay
A’s query. Then D selects A in the same way, which
leads to a loop query path: A�C�D�A. Thus, only
three nodes are queried in the whole query process
while all other nodes are invisible from the query
source A. If we change the metric to be the smallest la-
tency, the problem still exists because another loop is
formed from source A, A�C�B�A. It is very possible
that the query cannot be satisfied in the loop. This prob-
lem can be less serious when the size of the query sub-
set increases, which will be discussed in Section 3.
�
���
�
��
�
�
�
���
�
���
�
��
�
��
��
��
�� ��
��
��
��
��
��
��
�
�
���
�
�
��
�
��
�
��
�
���
�
��
�
���
�
��
�
��
�� �
��
���
��
��
��
��
�
�
��
�
(a) Query path loops (b) Non-optimal query path
Figure 1. The partial coverage problem
Many statistics-based search approaches use only
one metric to collect statistics information to select re-
lay neighbors, which does not always lead to an optimal
search path. Figure 1(b) shows an example in which A
is still the source node. When the search metric is the
volume of shared data, the query path would be
A�D�E along which the query will check 250 files in
200 unit of time. But obviously if the query path is
A�C�G�F�H, the query can check 500 files in 20
units of time. The first path selected using one search
metric is not as good as the second one.
3 Hybrid Periodical Flooding
In order to effectively reduce the traffic incurred by
flooding-based search and alleviate the partial coverage
problem, we propose Hybrid Periodical Flooding (HPF).
Before discuss HPF, we first define Periodical Flooding.
3.1 Periodical flooding (PF)
We notice that in all the existing statistics-based
search techniques, the number of relay neighbors, h,
does not change at all peers along the query path. In the
case of blind flooding, the phenomenon exhibits traffic
explosion. The concept of periodical flooding tries to
control the number of relay neighbors based on the TTL
value along the query path. More specifically, given a
peer with n logical neighbors and the current value of
TTL, the number of relay neighbors, h, is defined by the
following function h=f(n,TTL). Thus, in blind flooding
(BFS), we have h=fBFS(n,TTL)=n.. In DFS, we have
h=fDFS(n,TTL)=1.
The function h=f(n,TTL) can be viewed as a periodi-
cal function that changes as TTL changes. We call a
search mechanism using a periodical function as peri-
odic flooding (PF), in which the query mechanism is
divided into several phases that are periodically re-
peated. We call the number of different repeated phases
as a cycle, C. In all existing statistics-based search tech-
niques, they all have a cycle of C=1, which are special
cases of PF. We can ask the following questions in or-
der to design an efficient search mechanism. In what
conditions does a search mechanism with C=1 behave
better than a search mechanism with C>1? What is the
optimal value of C in terms of a desired performance
metric under different underlying physical network to-
pologies? For a given C, what is the optimal number of
relay neighbors? One example of PF functions with
C=2 is shown below:
�
�
�
�
�
�
�
��
�
��
�
��
�
��
�
�
evenisTTLifn
oddisTTLifn
TTLnf
,
3
1
,
2
1
),(
�� � ��
�
�
�
�
�
�
�
�
�
�
�� � ��
�
�
�
�
�
�
�
�
�
�
�������
��������
(a) BFS (b) PF
Figure 2. Comparison between BFS and a PF
We compare BFS and the example PF in Fig. 2. Sup-
pose peer O initiates a query. Blind flooding (BFS) is
employed in Fig. 2(a) where the query is sent or for-
warded 36 times to reach all the nodes. We use thin
connections to represent the links on which the query
traverses once and thick connections to represent the
links on which the query traverses twice. We have ex-
plained that for each query, each peer records the
neighbors, which forward the query to it. Thereby on
each link, at most two query messages can be sent
across it. When a link is traversed twice, the unneces-
sary traffic is incurred. For example, one of the mes-
sages from A to B and from B to A is unnecessary.
These redundant messages are shown in Fig. 2(a) using
dotted arrows.
Figure 2(b) illustrates the query process of the ex-
ample PF. Peer O has 4 neighbors and has TTL=7. We
randomly select relay neighbors. Peer O will select 2
nodes (that is n/2=2 since TTL=7 that is odd), peers A
and C, as relay neighbors. Peer A has 5 neighbors. It
will select 2 neighbors (G and I) to relay the query initi-
ated from peer O since TTL=6 and h=�n/3�=2. Simi-
larly, peer C relays the query to peer B and N (TTL=6
and h=�n/3�=2). Although the redundancy problem
still exists in PF (such as the traffics from B to J and
from I to J), it is significantly reduced compared with
that of BFS.
Table 1. PF and Blind Flooding
TTL Query Msg New Peers Msg Per Peer
7 4 4 1.00
6 17 8 2.12
BFS
5 15 2 7.50
7 2 2 1.00
6 4 4 1.00
PF
5 9 8 1.12
Table 1 compares the redundancy degree of both PF
and BFS. It presents the query messages relayed to new
peers. For example, in BFS, peers with TTL=5 relay the
query to 15 peers, but only 2 of the 15 peers receive the
query first time. In PF, peers with TTL=5 relay the
query to 9 peer of which 8 are first time receivers. That
means for peers with TTL=5, BFS sends 7.5 queries to
one new queried peer in average, while PF only sends
1.12 queries to one new queried peer in average. An ef-
ficient mechanism should query more peers using less
messages. Thus PF is much more efficient than BFS in
terms of traffic volume.
3.2 Hybrid Periodical Flooding
HPF Overview
After determining the number of relay neighbors (h),
a peer decides which h nodes should be selected. A
simple approach called Random Periodical Flooding
(RPF) selects h relay neighbors at random. Selecting re-
lay neighbors more objectively may result in better per-
formance. For example, we may use the shared data
volume as a metric to select query neighbors if we find
that peers with more shared data are more likely to sat-
isfy queries. By selecting the neighbors with larger
number of shard data, a query is more likely to succeed
in less number of hops than that of random selection.
We may also use the latency between the peer and its
neighbors as a metric to select neighbors. In this case,
for a given TTL value, a query will experience a shorter
delay. If we consider multiple metrics in relay neighbor
selection, the search mechanism is expected to have
better performance. This motivates us to propose Hy-
brid Periodical Flooding (HPF) in which the number of
relay neighbors can be changed periodically based on a
periodical function and the relay neighbors are selected
based on multiple metrics in a hybrid way.
HPF differentiates with RPF in that RPF selects re-
lay neighbors randomly, and differentiates with DBFS
in that DBFS only uses one metric to select relay
neighbors. HPF selects neighbors based on multiple
metrics and provides flexibility to justify different pa-
rameters to improve overall performance. Let h denote
the expected number of relay neighbors, which is given
by h = h1 + h2 + … + ht, where t is the number of met-
rics used in relay neighbor selection and hi is the num-
ber of relay neighbors selected by metric i.
Metrics
There are many metrics that may be used to select re-
lay neighbors, such as communication cost, bandwidth,
number of returned results from the neighbor, average
number of hops from the neighbor to peers who re-
sponded the previous queries, and so on. These metrics
may have different weights for a system with different
query access patterns or different performance require-
ments. For example, we may give higher weights to
some metrics that are more sensitive to the performance
in a specific system. We have ��
�
t
i
iw
1
1 , where iw is
the weight assigned to metric i ( ti ��1 ). To alleviate
the partial coverage problem, we select relay neighbors
in a hybrid way. We select hi neighbors using metric i,
where hi is determined by � �ii whh �� . Let Si denote
the set of neighbors selected based on the metric i. The
complete set of relay neighbors is i
t
i
SS
1�
� � , where
|S| iih � . Note that a neighbor may be selected by
more than one metric. Thus, the actual number of relay
neighbors selected may be less than h.
Termination of Search Queries
A query process is terminated when a pre-set TTL
value has been decreased to zero. Choosing an appro-
priate TTL value is very difficult. A large TTL may
cause higher traffic volume, while a small TTL may not
respond with enough number of query results. Further-
more there are no mutual feedbacks between the source
peer and the peers who forward or respond the query.
Thus it is hard for peers to know when to stop forward-
ing the query before the TTL value is reduced to zero.
Iterative Deepening [17] made an effort to address
this problem in some degree. In Iterative Deepening, a
policy P is used to control the search mechanism, which
provides a sequence of TTLs so that a query is flooded
from a very small TTL, and if necessary, to a gradually
enlarged scope. For example, one policy can be P={a, b,
c}, where P has three iterations. A query starts to be
flooded with TTL=a. If the query cannot be satisfied, it
will be flooded with TTL=b-a from all peers that are a
hops away from the source peer. Similarly if the query
still cannot be satisfied, it will be flooded with TTL=c-b
from all peers that are b hops away from the source peer.
In this policy, c is the maximal length of a query path.
Iterative Deepening is a good mechanism in the sense
that it alleviates the process time of middle nodes be-
tween iterations.
In HPF, we use this policy to terminate the success-
ful queries without incurring too much unnecessary
traffic. Since the combination is quite straightforward
and the performance of Iterative Deepening policy has
been evaluated in [17], this policy will not be re-
evaluated in this paper.
4 Simulation Methodology
We use simulation to evaluate the performance of
RPF and HPF and analyze the effects of the parameters.
4.1 Topology Generation
Two types of topologies, physical topology and logi-
cal topology, have to be generated in our simulation.
The physical topology should represent the real topol-
ogy with Internet characteristics. The logical topology
represents the overlay P2P topology built on top of the
physical topology. All P2P nodes are in the node subset
of the physical topology. The communication cost be-
tween two logical neighbors is calculated based on the
physical shortest path between this pair of nodes. To
simulate the performance of different search mecha-
nisms in a more realistic environment, the two topolo-
gies must accurately reflect the topological properties of
real networks in each layer.
Previous studies have shown that both large scale
Internet physical topologies [15] and P2P overlay to-
pologies follow small world and power law properties.
Power law describes the node degree while small world
describes characteristics of path length and clustering
coefficient [2]. Studies in [12] found that the topologies
generated using the AS Model have the properties of
small world and power law. BRITE [1] is a topology
generation tool that provides the option to generate to-
pologies based on the AS Model. Using BRITE, we
generate 10 physical topologies each with 10,000 nodes.
The logical topologies are generated with the number of
peers ranging from 1,000 to 5,000. The average number
of edges of each node is ranging from 6 to 20.
4.2 Simulation Setup
The total network traffic incurred by queries and av-
erage response time of all queries are two major metrics
that we use to evaluate the efficiency of a search
mechanism. High traffic volume will limit system scal-
ability and long response time is intolerable for users.
Network administrators care more about how much
network bandwidth consumed by a P2P system, while
users care more about the response time of queries,
which is viewed as a part of service quality of the sys-
tem.
0 5 10 15 20 25 30 35 40 45 50
0
1%
2%
3%
4%
5%
6%
7%
8%
Coverage Size
NodesDistribution(%)
1,000-node overlay network
10,000-node physical network
400 410 420 430 440 450 460 470 480 490 500
0
1%
2%
3%
4%
5%
6%
7%
8%
9%
NodesDistribution(%)
Coverage Size
1,000-node overlay network
10,000-node physical network
0 10 20 30 40 50 60 70 80 90 100
0
0.5%
1.0%
1.5%
2.0%
2.5%
3.0%
3.5%
4.0%
NodesDistribution(%)
Coverage Size
1,000-node overlay network
10,000-node physical network
Figure 3. Node distribution vs. cover-
age size. (h=1, metric 2)
Figure 4. Node distribution vs. cover-
age size. (h=2, metric 2)
Figure 5. Node distribution vs. cover-
age size. (h=1, metric 1)
In our simulation, we consider two metrics with the
same weight to select relay neighbors in HPF. In prac-
tice, more metrics could be used for neighbor selection.
The two metrics are the communication cost (metric 1)
that is the distance between a peer and its neighbor and
the shared number of files (metric 2) on each node.
Based on the first metric, a peer will select the
neighbors with the less communication costs. Based on
the second metric, a peer will select the neighbors with
the larger amount of shared data.
For each given search criterion, we distribute 100
files satisfying the search on the peers in a generated
P2P topology. That means there are totally 100 possible
results for a specific query in the whole P2P network.
The distribution of the 100 files on the network is ran-
dom. For each peer, we generate a number within 1 to
1000 as the number of shared files in this peer. Based
on the second metric in selecting relay neighbors, a
neighbor with more shared files is more likely to return
a response than a neighbor with less shared files.
5 Performance Evaluation
In this section, we present the simulation results to
show the effectiveness of HPF compared with DBFS
and BFS.
5.1 Partial Coverage Problem
Based on [3, 17], statistics-based search mechanisms
are more efficient and incur less traffic to the Internet
compared with blind flooding. However, statistics-
based search mechanisms have partial coverage prob-
lem as we discussed in Section 2.4. We quantitatively
illustrate the partial coverage problem in this section.
We first illustrate the case in which only one relay
neighbor is selected to send/forward a query (h=1)
based on the number of shared files in neighbors. We
set TTL as infinity. Figure 3 shows the node distribution
versus the number of peers being queried, which is de-
fined as coverage size. For example, queries initiated
from 8% of peers can only reach 10 other peers. Most
of peers can only push their queries to 10 to 30 other
peers. This means that loops are formed and only a very
small number of peers can be reached for any queries.
Note that the overlay network has 1000 nodes and the
physical network has 10,000 nodes. Figure 4 illustrates
the node distribution versus the coverage size, where
h=2 and TTL=infinity. The coverage size is about 400
peers in average, which is still a small number in a P2P
network.
Figure 5 shows node distribution versus coverage
size when we use network latency as the metric to se-
lect relay neighbors. Again, we see the partial coverage
problem. The partial coverage problem will disappear
when h=n, which is the case of blind flooding. We did
the same group of simulations on different topologies
using different metrics. The results are quite consistent.
Figure 6 shows the percentage of covered peers to total
peers versus the number of relay neighbors (h=1, 2, n/5,
n/4, n/3, n/2, and Sqrt(n)). The percentage of coverage
is larger for a larger h. A larger h means a smaller
chance for all reached peers to form a loop.
One Two 1/5 1/4 1/3 Half Sqrt
0
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Number of Relay Neighbors
PercentageofCoveredNodes
1.65%
42.4%
56.3%
63.0%
77.5%
87.5%
77.6%
Figure 6. Percentage of coverage vs. the number of
relay neighbors
5.2 Performance of Random PF
We have evaluated network traffic and average re-
sponse time of RPF that selects relay neighbors at ran-
dom. We can use many different periodical flooding
functions to determine the number of relay neighbors.
These functions should not be over complicated. We
have tried tens of periodical flooding functions with dif-
ferent C.
0 5 10 15 20 25 30 35 40
0
10
20
30
40
50
60
70
80
90
100
Requested Number of Response Results
NormalizedQueryCost
BFS
RPF (1)
RPF (2)
RPF (3)
0 5 10 15 20 25 30 35 40
0
10
20
30
40
50
60
70
80
90
100
Requested Number of Response Results
NormalizedResponseTime
BFS
RPF (1)
RPF (2)
RPF (3)
DFS
0 5 10 15 20 25 30 35 40
0
10
20
30
40
50
60
70
80
90
100
Requested Number of Response Results
NormalizedQueryCost
BFS
RPF
DBFS
HPF
Figure 7. Normalized traffic of RPF Figure 8. Normalized response
time of RPF
Figure 9. Normalized traffic com-
parison
Figures 7 and 8 show the normalized network traffic
cost and normalized average response time versus the
required number of response results. The traffic and av-
erage response time always perform in opposite way. If
a search mechanism causes low traffic, it will suffer
from high response time and vice versa. RPF is de-
signed to provide an opportunity to have a tradeoff be-
tween total traffic and average response time, thus ob-
taining a better overall search performance. We may
expect a search mechanism to reduce a large amount of
traffic by increasing a little more response time or vice
versa. How to quantitatively measure the overall per-
formance based on the tradeoff is an issue.
It’s hard to find the best search mechanism. We de-
fine p to measure the overall performance, where
timetrafficp RC �� �� , traffic and time are nor-
malized value of total network traffic and average re-
sponse time, �C and �R are the weight parameters for
network traffic and response time, and �C + �R = 1. We
seek an asymptotically periodical flooding function
fa(n,TTL) such that p can be minimal or close to mini-
mal. If a system emphasizes more on low network traf-
fic, we can set �C > �R; otherwise, we can set �C < �R
for a system emphasizing more on quick response time.
Based on different topologies with different number
of average connections, and different values of �C and
�R, the functions of fa(n,TTL) may be derived differently.
In our simulation of HPF, the average number of edge
connections is 10. We choose �C = 0.6 and �R = 0.4.
Thus, the corresponding period function is derived as:
�
�
�
�
�
�
�
��
�
��
�
��
�
��
�
�
evenisTTLifn
oddisTTLifn
TTLnf
,
4
1
,
2
1
),(
5.3 Effectiveness of HPF
HPF selects relay neighbors based on multiple met-
rics in a hybrid way. We use communication cost and
the volume of shared data as two metrics to select relay
neighbors.
Based on the simulation over 10,000 queries, Figure
9 shows the normalized network traffic versus the re-
quired number of response results of four different
search mechanisms: BFS, RPF, DBFS and HPF. DBFS
reduces the network traffic by 30~50% compared with
BFS. HPF outperforms DBFS by up to 20%. Figure 10
compares the normalized response time of four different
search mechanisms over 10,000 queries versus the re-
quired number of response results. HPF performs the
best compared with RPF and DBFS, but still worse than
BFS. DBFS selects relay neighbors who have the larg-
est volume of shared files. Each query may get more re-
sults by reaching fewer peers. HPF needs to query more
peers to obtain the same amount of results than DBFS
but much less than BFS and RPF. That is because we
use multiple metrics instead of a single metric used in
DBFS, expecting to obtain better overall performance,
which has been shown in Figs. 9 and 10.
5.4 Alleviating the Partial Coverage Problem
HPF can effectively address the partial coverage
problem discussed in Section 2.4. Figure 11 shows the
percentage of queried peers as TTL increases. BFS can
quickly cover 100% peers, while DBFS can only cover
up to 77% peers in our simulation because of the partial
coverage problem. DBFS still covers only around 77%
when the value of TTL is set to infinity in our simula-
tion. However, HPF and RPF can cover more than 96%
peers as TTL is increased to 10.
Figure 12 compares the peer coverage size of DBFS
and HPF. In DBFS, most nodes can cover 760-780
peers out of 1,000 nodes. The coverage size is increased
to 950-970 in HPF.
0 5 10 15 20 25 30 35 40
0
10
20
30
40
50
60
70
80
90
100
Requested Number of Response Results
NormallizedResponseTime
BFS
RPF
DBFS
HPF
0 2 4 6 8 10 12 14
0
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
TTL
PercentageofCoveredNodes
BFS
HPF
DBFS
RPF
600 650 700 750 800 850 900 950 1000
0
2%
4%
6%
8%
10%
12%
14%
16%
18%
Coverage Size
NodesDistribution
DBFS
HPF
Figure 10. Normalized response
time comparison
Figure 11. Coverage percentage
comparison
Figure 12. Partial coverage
comparison
6 Conclusion
In this paper, we have proposed an efficient and
adaptive search mechanism, Hybrid Periodical Flooding.
HPF improves the efficiency of blind flooding by re-
taining the advantages of statistics-based search mecha-
nisms and by alleviating the partial coverage problem.
We summarize our contributions as follows:
� Analyze the current search mechanisms used and
proposed in unstructured P2P networks.
� Qualitatively and quantitatively analyze the partial
coverage problem caused by statistics-based search
mechanisms, such as DBFS.
� Propose to use a periodical flooding function to de-
fine the number of relay neighbors, which can be
adaptively changed. This is the first technique used
in HPF.
� Propose to use multiple metrics to select relay
neighbors to obtain better overall performance or
adaptively meet different performance requirements,
which is the second technique used in HPF.
We have shown the performance of HPF using two
metrics to select relay neighbors. HPF provides the
flexibility to use more metrics and allows the applica-
tion to define multiple metrics and give them different
weights, thereby the algorithm is more flexible in prac-
tice to meet different performance requirements.
References
[1] BRITE, http://www.cs.bu.edu/brite/.
[2] T. Bu and D. Towsley, On distinguishing between Inter-
net power law topology generators, In Proceedings of
IEEE INFOCOM'02 Conference, 2002.
[3] A. Crespo and H. Garcia-Molina, Routing indices for
peer-to-peer systems, In Proceedings of 22nd Interna-
tional Conference on Distributed Computing Systems,
2002.
[4] E.Cohen, A.Fiat, and H.Kaplan, Associative search in
peer to peer networks: harnessing latent semantics, In
Proceedings of the IEEE INFOCOM'03, 2003.
[5] Fasttrack, http://www.fasttrack.nu/.
[6] Freenet, http://freenet.sourceforge.net.
[7] Gnutella, http://gnutella.wego.com/.
[8] KaZaA, http://www.kazaa.com.
[9] B. Krishnamurthy and J. Wang, Automated traffic classi-
fication for application-specific peering, In Proceedings
of ACM SIGCOMM Internet Measurement Workshop,
November 2002.
[10] Q. Lv, et al., Search and replication in unstructured peer-
to-peer networks, In Proceedings of the 16th ACM Inter-
national Conference on Supercomputing, 2002.
[11] Ritter, Why Gnutella can't scale. No, really.
http://www.tch.org/gnutella.html.
[12] S. Saroiu, P. Gummadi, and S. Gribble, A measurement
study of peer-to-peer file sharing systems, In Proceedings
of Multimedia Computing and Networking (MMCN),
2002.
[13] S. Sen and J. Wang, Analyzing peer-to-peer traffic
across large networks, In Proceedings of ACM SIG-
COMM Internet Measurement Workshop, 2002.
[14] K. Sripanidkulchai, B. Maggs, and H. Zhang, Efficient
content location using interest-based locality in peer-to-
peer systems, In Proceedings of INFOCOM'03, 2003.
[15] H. Tangmunarunkit, et al., Network topology generators:
degree-based vs. structural, In Proceedings of In Pro-
ceedings of SIGCOMM'02, 2002.
[16] B. Yang and H. Garcia-Molina, Designing a super-peer
network, In Proceedings of the 19th International Con-
ference on Data Engineering (ICDE), March 2003.
[17] B. Yang and H. Garcia-Molina, Efficient search in peer-
to-peer networks, In Proceedings of ICDCS'02, 2002.

More Related Content

What's hot

An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...
An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...
An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...IRJET Journal
 
A Proximity-Aware Interest-Clustered P2P File Sharing System
A Proximity-Aware Interest-Clustered P2P File Sharing System A Proximity-Aware Interest-Clustered P2P File Sharing System
A Proximity-Aware Interest-Clustered P2P File Sharing System 1crore projects
 
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe System
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe SystemOntology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe System
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe Systemtheijes
 
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...1crore projects
 
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTION
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTIONIEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTION
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTIONranjith kumar
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER) International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER) ijceronline
 
Study of the topology mismatch problem in peer to-peer networks
Study of the topology mismatch problem in peer to-peer networksStudy of the topology mismatch problem in peer to-peer networks
Study of the topology mismatch problem in peer to-peer networksAlexander Decker
 
Comprehensive Path Quality Measurement in Wireless Sensor Networks
Comprehensive Path Quality Measurement in Wireless Sensor NetworksComprehensive Path Quality Measurement in Wireless Sensor Networks
Comprehensive Path Quality Measurement in Wireless Sensor NetworksIJTET Journal
 
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...Developing QoS by Priority Routing for Real Time Data in Internet of Things (...
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...IJCNCJournal
 
Effective Data Retrieval System with Bloom in a Unstructured p2p Network
Effective Data Retrieval System with Bloom in a Unstructured p2p NetworkEffective Data Retrieval System with Bloom in a Unstructured p2p Network
Effective Data Retrieval System with Bloom in a Unstructured p2p NetworkUvaraj Shan
 

What's hot (10)

An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...
An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...
An Extensive Literature Review of Various Routing Protocols in Delay Tolerant...
 
A Proximity-Aware Interest-Clustered P2P File Sharing System
A Proximity-Aware Interest-Clustered P2P File Sharing System A Proximity-Aware Interest-Clustered P2P File Sharing System
A Proximity-Aware Interest-Clustered P2P File Sharing System
 
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe System
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe SystemOntology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe System
Ontology-Based Routing for Large-Scale Unstructured P2P Publish/Subscribe System
 
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...
Maximizing P2P File Access Availability in Mobile Ad Hoc Networks though Repl...
 
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTION
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTIONIEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTION
IEEE BE-BTECH NS2 PROJECT@ DREAMWEB TECHNO SOLUTION
 
International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER) International Journal of Computational Engineering Research(IJCER)
International Journal of Computational Engineering Research(IJCER)
 
Study of the topology mismatch problem in peer to-peer networks
Study of the topology mismatch problem in peer to-peer networksStudy of the topology mismatch problem in peer to-peer networks
Study of the topology mismatch problem in peer to-peer networks
 
Comprehensive Path Quality Measurement in Wireless Sensor Networks
Comprehensive Path Quality Measurement in Wireless Sensor NetworksComprehensive Path Quality Measurement in Wireless Sensor Networks
Comprehensive Path Quality Measurement in Wireless Sensor Networks
 
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...Developing QoS by Priority Routing for Real Time Data in Internet of Things (...
Developing QoS by Priority Routing for Real Time Data in Internet of Things (...
 
Effective Data Retrieval System with Bloom in a Unstructured p2p Network
Effective Data Retrieval System with Bloom in a Unstructured p2p NetworkEffective Data Retrieval System with Bloom in a Unstructured p2p Network
Effective Data Retrieval System with Bloom in a Unstructured p2p Network
 

Viewers also liked

Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...
Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...
Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...Zhenyun Zhuang
 
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...Zhenyun Zhuang
 
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...Building Cloud-ready Video Transcoding System for Content Delivery Networks (...
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...Zhenyun Zhuang
 
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...Zhenyun Zhuang
 
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...Zhenyun Zhuang
 
Mutual Exclusion in Wireless Sensor and Actor Networks
Mutual Exclusion in Wireless Sensor and Actor NetworksMutual Exclusion in Wireless Sensor and Actor Networks
Mutual Exclusion in Wireless Sensor and Actor NetworksZhenyun Zhuang
 
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...Zhenyun Zhuang
 
Designing SSD-friendly Applications for Better Application Performance and Hi...
Designing SSD-friendly Applications for Better Application Performance and Hi...Designing SSD-friendly Applications for Better Application Performance and Hi...
Designing SSD-friendly Applications for Better Application Performance and Hi...Zhenyun Zhuang
 
Improving energy efficiency of location sensing on smartphones
Improving energy efficiency of location sensing on smartphonesImproving energy efficiency of location sensing on smartphones
Improving energy efficiency of location sensing on smartphonesZhenyun Zhuang
 
Dynamic Layer Management in Super-Peer Architectures
Dynamic Layer Management in Super-Peer ArchitecturesDynamic Layer Management in Super-Peer Architectures
Dynamic Layer Management in Super-Peer ArchitecturesZhenyun Zhuang
 
Capacity Planning and Headroom Analysis for Taming Database Replication Latency
Capacity Planning and Headroom Analysis for Taming Database Replication LatencyCapacity Planning and Headroom Analysis for Taming Database Replication Latency
Capacity Planning and Headroom Analysis for Taming Database Replication LatencyZhenyun Zhuang
 
Wireless memory: Eliminating communication redundancy in Wi-Fi networks
Wireless memory: Eliminating communication redundancy in Wi-Fi networksWireless memory: Eliminating communication redundancy in Wi-Fi networks
Wireless memory: Eliminating communication redundancy in Wi-Fi networksZhenyun Zhuang
 
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...Application-Aware Acceleration for Wireless Data Networks: Design Elements an...
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...Zhenyun Zhuang
 
AOTO: Adaptive overlay topology optimization in unstructured P2P systems
AOTO: Adaptive overlay topology optimization in unstructured P2P systemsAOTO: Adaptive overlay topology optimization in unstructured P2P systems
AOTO: Adaptive overlay topology optimization in unstructured P2P systemsZhenyun Zhuang
 
Hazard avoidance in wireless sensor and actor networks
Hazard avoidance in wireless sensor and actor networksHazard avoidance in wireless sensor and actor networks
Hazard avoidance in wireless sensor and actor networksZhenyun Zhuang
 
Optimizing JMS Performance for Cloud-based Application Servers
Optimizing JMS Performance for Cloud-based Application ServersOptimizing JMS Performance for Cloud-based Application Servers
Optimizing JMS Performance for Cloud-based Application ServersZhenyun Zhuang
 
A3: application-aware acceleration for wireless data networks
A3: application-aware acceleration for wireless data networksA3: application-aware acceleration for wireless data networks
A3: application-aware acceleration for wireless data networksZhenyun Zhuang
 
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified Worms
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified WormsPAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified Worms
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified WormsZhenyun Zhuang
 
WebAccel: Accelerating Web access for low-bandwidth hosts
WebAccel: Accelerating Web access for low-bandwidth hostsWebAccel: Accelerating Web access for low-bandwidth hosts
WebAccel: Accelerating Web access for low-bandwidth hostsZhenyun Zhuang
 
Optimizing Streaming Server Selection for CDN-delivered Live Streaming
Optimizing Streaming Server Selection for CDN-delivered Live StreamingOptimizing Streaming Server Selection for CDN-delivered Live Streaming
Optimizing Streaming Server Selection for CDN-delivered Live StreamingZhenyun Zhuang
 

Viewers also liked (20)

Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...
Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...
Mobile Hosts Participating in Peer-to-Peer Data Networks: Challenges and Solu...
 
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...
Eliminating OS-caused Large JVM Pauses for Latency-sensitive Java-based Cloud...
 
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...Building Cloud-ready Video Transcoding System for Content Delivery Networks (...
Building Cloud-ready Video Transcoding System for Content Delivery Networks (...
 
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...
Optimizing CDN Infrastructure for Live Streaming with Constrained Server Chai...
 
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...
OCPA: An Algorithm for Fast and Effective Virtual Machine Placement and Assig...
 
Mutual Exclusion in Wireless Sensor and Actor Networks
Mutual Exclusion in Wireless Sensor and Actor NetworksMutual Exclusion in Wireless Sensor and Actor Networks
Mutual Exclusion in Wireless Sensor and Actor Networks
 
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...
Guarding Fast Data Delivery in Cloud: an Effective Approach to Isolating Perf...
 
Designing SSD-friendly Applications for Better Application Performance and Hi...
Designing SSD-friendly Applications for Better Application Performance and Hi...Designing SSD-friendly Applications for Better Application Performance and Hi...
Designing SSD-friendly Applications for Better Application Performance and Hi...
 
Improving energy efficiency of location sensing on smartphones
Improving energy efficiency of location sensing on smartphonesImproving energy efficiency of location sensing on smartphones
Improving energy efficiency of location sensing on smartphones
 
Dynamic Layer Management in Super-Peer Architectures
Dynamic Layer Management in Super-Peer ArchitecturesDynamic Layer Management in Super-Peer Architectures
Dynamic Layer Management in Super-Peer Architectures
 
Capacity Planning and Headroom Analysis for Taming Database Replication Latency
Capacity Planning and Headroom Analysis for Taming Database Replication LatencyCapacity Planning and Headroom Analysis for Taming Database Replication Latency
Capacity Planning and Headroom Analysis for Taming Database Replication Latency
 
Wireless memory: Eliminating communication redundancy in Wi-Fi networks
Wireless memory: Eliminating communication redundancy in Wi-Fi networksWireless memory: Eliminating communication redundancy in Wi-Fi networks
Wireless memory: Eliminating communication redundancy in Wi-Fi networks
 
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...Application-Aware Acceleration for Wireless Data Networks: Design Elements an...
Application-Aware Acceleration for Wireless Data Networks: Design Elements an...
 
AOTO: Adaptive overlay topology optimization in unstructured P2P systems
AOTO: Adaptive overlay topology optimization in unstructured P2P systemsAOTO: Adaptive overlay topology optimization in unstructured P2P systems
AOTO: Adaptive overlay topology optimization in unstructured P2P systems
 
Hazard avoidance in wireless sensor and actor networks
Hazard avoidance in wireless sensor and actor networksHazard avoidance in wireless sensor and actor networks
Hazard avoidance in wireless sensor and actor networks
 
Optimizing JMS Performance for Cloud-based Application Servers
Optimizing JMS Performance for Cloud-based Application ServersOptimizing JMS Performance for Cloud-based Application Servers
Optimizing JMS Performance for Cloud-based Application Servers
 
A3: application-aware acceleration for wireless data networks
A3: application-aware acceleration for wireless data networksA3: application-aware acceleration for wireless data networks
A3: application-aware acceleration for wireless data networks
 
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified Worms
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified WormsPAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified Worms
PAIDS: A Proximity-Assisted Intrusion Detection System for Unidentified Worms
 
WebAccel: Accelerating Web access for low-bandwidth hosts
WebAccel: Accelerating Web access for low-bandwidth hostsWebAccel: Accelerating Web access for low-bandwidth hosts
WebAccel: Accelerating Web access for low-bandwidth hosts
 
Optimizing Streaming Server Selection for CDN-delivered Live Streaming
Optimizing Streaming Server Selection for CDN-delivered Live StreamingOptimizing Streaming Server Selection for CDN-delivered Live Streaming
Optimizing Streaming Server Selection for CDN-delivered Live Streaming
 

Similar to HPF

A Distributed Approach to Solving Overlay Mismatching Problem
A Distributed Approach to Solving Overlay Mismatching ProblemA Distributed Approach to Solving Overlay Mismatching Problem
A Distributed Approach to Solving Overlay Mismatching ProblemZhenyun Zhuang
 
Research Inventy : International Journal of Engineering and Science
Research Inventy : International Journal of Engineering and ScienceResearch Inventy : International Journal of Engineering and Science
Research Inventy : International Journal of Engineering and Scienceinventy
 
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...IEEEGLOBALSOFTTECHNOLOGIES
 
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERS
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERSA QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERS
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERSijait
 
Vol 11 No 1 - March 2014
Vol 11 No 1 - March 2014Vol 11 No 1 - March 2014
Vol 11 No 1 - March 2014ijcsbi
 
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...inventionjournals
 
Flexible bloom for searching textual content
Flexible bloom for searching textual contentFlexible bloom for searching textual content
Flexible bloom for searching textual contentUvaraj Shan
 
Flexible bloom for searching textual content
Flexible bloom for searching textual contentFlexible bloom for searching textual content
Flexible bloom for searching textual contentUvaraj Shan
 
Efficient Destination Discovery using Geographical Gossiping in MANETs
Efficient Destination Discovery using Geographical Gossiping in MANETsEfficient Destination Discovery using Geographical Gossiping in MANETs
Efficient Destination Discovery using Geographical Gossiping in MANETsidescitation
 
Token Based Packet Loss Control Mechanism for Networks
Token Based Packet Loss Control Mechanism for NetworksToken Based Packet Loss Control Mechanism for Networks
Token Based Packet Loss Control Mechanism for NetworksIJMER
 
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMSEFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMSijp2p
 
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS ijp2p
 
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSN
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSNDistributed Three Hop Routing Protocol for Enhancing Routing Process in WSN
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSNpaperpublications3
 
Study on security and quality of service implementations in p2 p overlay netw...
Study on security and quality of service implementations in p2 p overlay netw...Study on security and quality of service implementations in p2 p overlay netw...
Study on security and quality of service implementations in p2 p overlay netw...eSAT Publishing House
 
A way of managing data center networks
A way of managing data center networksA way of managing data center networks
A way of managing data center networksIOSR Journals
 
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...journalBEEI
 

Similar to HPF (20)

Optimizing of Bloom Filters by Automatic Bloom Filter Updating and Instantly...
Optimizing of Bloom Filters by Automatic Bloom Filter Updating  and Instantly...Optimizing of Bloom Filters by Automatic Bloom Filter Updating  and Instantly...
Optimizing of Bloom Filters by Automatic Bloom Filter Updating and Instantly...
 
A Distributed Approach to Solving Overlay Mismatching Problem
A Distributed Approach to Solving Overlay Mismatching ProblemA Distributed Approach to Solving Overlay Mismatching Problem
A Distributed Approach to Solving Overlay Mismatching Problem
 
Research Inventy : International Journal of Engineering and Science
Research Inventy : International Journal of Engineering and ScienceResearch Inventy : International Journal of Engineering and Science
Research Inventy : International Journal of Engineering and Science
 
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...
JAVA 2013 IEEE NETWORKING PROJECT Transfer reliability and congestion control...
 
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERS
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERSA QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERS
A QUERY LEARNING ROUTING APPROACH BASED ON SEMANTIC CLUSTERS
 
Fu2510631066
Fu2510631066Fu2510631066
Fu2510631066
 
A Comparison between Flooding and Bloom Filter Based Multikeyword Search in P...
A Comparison between Flooding and Bloom Filter Based Multikeyword Search in P...A Comparison between Flooding and Bloom Filter Based Multikeyword Search in P...
A Comparison between Flooding and Bloom Filter Based Multikeyword Search in P...
 
Vol 11 No 1 - March 2014
Vol 11 No 1 - March 2014Vol 11 No 1 - March 2014
Vol 11 No 1 - March 2014
 
Congestion control mechanism using network border protocol
Congestion control mechanism using network border protocolCongestion control mechanism using network border protocol
Congestion control mechanism using network border protocol
 
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...
Link Prediction And Link Establishment Based On Network Nodes Life Time In Mo...
 
Flexible bloom for searching textual content
Flexible bloom for searching textual contentFlexible bloom for searching textual content
Flexible bloom for searching textual content
 
Flexible bloom for searching textual content
Flexible bloom for searching textual contentFlexible bloom for searching textual content
Flexible bloom for searching textual content
 
Efficient Destination Discovery using Geographical Gossiping in MANETs
Efficient Destination Discovery using Geographical Gossiping in MANETsEfficient Destination Discovery using Geographical Gossiping in MANETs
Efficient Destination Discovery using Geographical Gossiping in MANETs
 
Token Based Packet Loss Control Mechanism for Networks
Token Based Packet Loss Control Mechanism for NetworksToken Based Packet Loss Control Mechanism for Networks
Token Based Packet Loss Control Mechanism for Networks
 
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMSEFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
 
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
EFFECTIVE TOPOLOGY-AWARE PEER SELECTION IN UNSTRUCTURED PEER-TO-PEER SYSTEMS
 
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSN
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSNDistributed Three Hop Routing Protocol for Enhancing Routing Process in WSN
Distributed Three Hop Routing Protocol for Enhancing Routing Process in WSN
 
Study on security and quality of service implementations in p2 p overlay netw...
Study on security and quality of service implementations in p2 p overlay netw...Study on security and quality of service implementations in p2 p overlay netw...
Study on security and quality of service implementations in p2 p overlay netw...
 
A way of managing data center networks
A way of managing data center networksA way of managing data center networks
A way of managing data center networks
 
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...
Routing performance of structured overlay in Distributed Hash Tables (DHT) fo...
 

More from Zhenyun Zhuang

Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...
Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...
Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...Zhenyun Zhuang
 
On the Impact of Mobile Hosts in Peer-to-Peer Data Networks
On the Impact of Mobile Hosts in Peer-to-Peer Data NetworksOn the Impact of Mobile Hosts in Peer-to-Peer Data Networks
On the Impact of Mobile Hosts in Peer-to-Peer Data NetworksZhenyun Zhuang
 
Client-side web acceleration for low-bandwidth hosts
Client-side web acceleration for low-bandwidth hostsClient-side web acceleration for low-bandwidth hosts
Client-side web acceleration for low-bandwidth hostsZhenyun Zhuang
 
Enhancing Intrusion Detection System with Proximity Information
Enhancing Intrusion Detection System with Proximity InformationEnhancing Intrusion Detection System with Proximity Information
Enhancing Intrusion Detection System with Proximity InformationZhenyun Zhuang
 
SLA-aware Dynamic CPU Scaling in Business Cloud Computing Environments
SLA-aware Dynamic CPU Scaling in Business Cloud Computing EnvironmentsSLA-aware Dynamic CPU Scaling in Business Cloud Computing Environments
SLA-aware Dynamic CPU Scaling in Business Cloud Computing EnvironmentsZhenyun Zhuang
 
OS caused Large JVM pauses: Deep dive and solutions
OS caused Large JVM pauses: Deep dive and solutionsOS caused Large JVM pauses: Deep dive and solutions
OS caused Large JVM pauses: Deep dive and solutionsZhenyun Zhuang
 
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...Ensuring High-performance of Mission-critical Java Applications in Multi-tena...
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...Zhenyun Zhuang
 

More from Zhenyun Zhuang (7)

Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...
Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...
Optimized Selection of Streaming Servers with GeoDNS for CDN Delivered Live S...
 
On the Impact of Mobile Hosts in Peer-to-Peer Data Networks
On the Impact of Mobile Hosts in Peer-to-Peer Data NetworksOn the Impact of Mobile Hosts in Peer-to-Peer Data Networks
On the Impact of Mobile Hosts in Peer-to-Peer Data Networks
 
Client-side web acceleration for low-bandwidth hosts
Client-side web acceleration for low-bandwidth hostsClient-side web acceleration for low-bandwidth hosts
Client-side web acceleration for low-bandwidth hosts
 
Enhancing Intrusion Detection System with Proximity Information
Enhancing Intrusion Detection System with Proximity InformationEnhancing Intrusion Detection System with Proximity Information
Enhancing Intrusion Detection System with Proximity Information
 
SLA-aware Dynamic CPU Scaling in Business Cloud Computing Environments
SLA-aware Dynamic CPU Scaling in Business Cloud Computing EnvironmentsSLA-aware Dynamic CPU Scaling in Business Cloud Computing Environments
SLA-aware Dynamic CPU Scaling in Business Cloud Computing Environments
 
OS caused Large JVM pauses: Deep dive and solutions
OS caused Large JVM pauses: Deep dive and solutionsOS caused Large JVM pauses: Deep dive and solutions
OS caused Large JVM pauses: Deep dive and solutions
 
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...Ensuring High-performance of Mission-critical Java Applications in Multi-tena...
Ensuring High-performance of Mission-critical Java Applications in Multi-tena...
 

Recently uploaded

High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsCall Girls in Nagpur High Profile
 
Biology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxBiology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxDeepakSakkari2
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )Tsuyoshi Horigome
 
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort servicejennyeacort
 
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSAPPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSKurinjimalarL3
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfAsst.prof M.Gokilavani
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerAnamika Sarkar
 
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdfCCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdfAsst.prof M.Gokilavani
 
GDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSCAESB
 
Software and Systems Engineering Standards: Verification and Validation of Sy...
Software and Systems Engineering Standards: Verification and Validation of Sy...Software and Systems Engineering Standards: Verification and Validation of Sy...
Software and Systems Engineering Standards: Verification and Validation of Sy...VICTOR MAESTRE RAMIREZ
 
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝soniya singh
 
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...ranjana rawat
 
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVRajaP95
 
Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024hassan khalil
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.eptoze12
 
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...ZTE
 
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptx
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptxDecoding Kotlin - Your guide to solving the mysterious in Kotlin.pptx
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptxJoão Esperancinha
 

Recently uploaded (20)

High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur EscortsHigh Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
High Profile Call Girls Nagpur Meera Call 7001035870 Meet With Nagpur Escorts
 
Biology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptxBiology for Computer Engineers Course Handout.pptx
Biology for Computer Engineers Course Handout.pptx
 
SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )SPICE PARK APR2024 ( 6,793 SPICE Models )
SPICE PARK APR2024 ( 6,793 SPICE Models )
 
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
VICTOR MAESTRE RAMIREZ - Planetary Defender on NASA's Double Asteroid Redirec...
 
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort serviceGurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
Gurgaon ✡️9711147426✨Call In girls Gurgaon Sector 51 escort service
 
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICSAPPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
APPLICATIONS-AC/DC DRIVES-OPERATING CHARACTERISTICS
 
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdfCCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
CCS355 Neural Network & Deep Learning Unit II Notes with Question bank .pdf
 
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
🔝9953056974🔝!!-YOUNG call girls in Rajendra Nagar Escort rvice Shot 2000 nigh...
 
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube ExchangerStudy on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
Study on Air-Water & Water-Water Heat Exchange in a Finned Tube Exchanger
 
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdfCCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
CCS355 Neural Network & Deep Learning UNIT III notes and Question bank .pdf
 
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Serviceyoung call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
young call girls in Rajiv Chowk🔝 9953056974 🔝 Delhi escort Service
 
GDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentationGDSC ASEB Gen AI study jams presentation
GDSC ASEB Gen AI study jams presentation
 
Software and Systems Engineering Standards: Verification and Validation of Sy...
Software and Systems Engineering Standards: Verification and Validation of Sy...Software and Systems Engineering Standards: Verification and Validation of Sy...
Software and Systems Engineering Standards: Verification and Validation of Sy...
 
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝
Model Call Girl in Narela Delhi reach out to us at 🔝8264348440🔝
 
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
(ANVI) Koregaon Park Call Girls Just Call 7001035870 [ Cash on Delivery ] Pun...
 
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IVHARMONY IN THE NATURE AND EXISTENCE - Unit-IV
HARMONY IN THE NATURE AND EXISTENCE - Unit-IV
 
Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024Architect Hassan Khalil Portfolio for 2024
Architect Hassan Khalil Portfolio for 2024
 
Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.Oxy acetylene welding presentation note.
Oxy acetylene welding presentation note.
 
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...
ZXCTN 5804 / ZTE PTN / ZTE POTN / ZTE 5804 PTN / ZTE POTN 5804 ( 100/200 GE Z...
 
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptx
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptxDecoding Kotlin - Your guide to solving the mysterious in Kotlin.pptx
Decoding Kotlin - Your guide to solving the mysterious in Kotlin.pptx
 

HPF

  • 1. Hybrid Periodical Flooding in Unstructured Peer-to-Peer Networks* * This work was partially supported by Michigan State University IRGP Grant 41114 and by Hong Kong RGC Grant HKUST6161/03E. Zhenyun Zhuang1 , Yunhao Liu1 , Li Xiao1 and Lionel M. Ni2 1 Department of Computer Science and Engineering, Michigan State University, U.S.A. 2 Department of Computer Science, Hong Kong University of Science and Technology, Hong Kong {zhuangz1, liuyunha, lxiao}@cse.msu.edu, ni@cs.ust.hk Abstract Blind flooding is a popular search mechanism used in current commercial P2P systems because of its sim- plicity. However, blind flooding among peers or super- peers causes large volume of unnecessary traffic al- though the response time is short. Some improved sta- tistics-based search mechanisms can reduce the traffic volume but also significantly shrink the query coverage range. In some search mechanisms, not all peers may be reachable creating the so-called partial coverage problem. Aiming at alleviating the partial coverage problem and reducing the unnecessary traffic, we pro- pose an efficient and adaptive search mechanism, Hy- brid Periodical Flooding (HPF). HPF retains the ad- vantages of statistics-based search mechanisms, allevi- ates the partial coverage problem, and provides the flexibility to adaptively adjust different parameters to meet different performance requirements. The effective- ness of HPF is demonstrated through simulation studies. 1 Introduction In an unstructured P2P system, such as Gnutella [7] and KaZaA [8], file placement is random, which has no correlation with the network topology [17]. Unstruc- tured P2P systems are most commonly used in today's Internet. In an unstructured P2P system, when a source peer needs to query an object, it sends a query to its neighbors. If a peer receiving the query cannot provide the requested object, it may relay the query to its own neighbors. If the peer receiving the query can provide the requested object, a response message will be sent back to the source peer along the inverse of the query path. The most popular query operation in use, such as Gnutella and KaZaA (among supernodes), is to blindly “flood" a query to the network. A query is broadcast and rebroadcast until a certain criterion is satisfied. This mechanism ensures that the query will be “flooded” to as many peers as possible within a short period of time in a P2P overlay network. However, flooding also causes a lot of network traffic and most of which is un- necessary. Study in [13] shows that P2P traffic contrib- utes the largest portion of the Internet traffic based on their measurements on three popular P2P systems, FastTrack (including KaZaA and Grokster) [5], Gnutella, and DirectConnect. The inefficient blind flooding search technique causes the unstructured P2P systems being far from scalable [11]. To avoid the large volume of unnecessary traffic in- curred by flooding-based search, many efforts have been made to improve search algorithms for unstruc- tured P2P systems. One typical approach is statistics- based, in which instead of flooding to all immediate overlay neighbors, a peer selects only a subset of its neighbors to query based on some statistics information of some metrics and heuristic algorithms. When han- dling a query message (either relayed from its neighbor or originated from itself) in a statistics-based search al- gorithm, the peer determines the subset of its logical neighbors to relay the query message. Statistics-based search mechanisms may significantly reduce the traffic volume but may also reduce the query coverage range so that a query may traverse a longer path to be satisfied or cannot be satisfied. In some search mechanisms, not all peers may be reachable creating the so-called partial coverage problem. Our objective is trying to alleviate the partial coverage problem and reduce unnecessary traffic. In this paper, Section 2 will give an overview and classification of known search mechanisms. The con- cept of our proposed periodical flooding method will be introduced in Section 3. Based on periodical flooding and weighted metrics in selecting relay neighbors, the hybrid periodical flooding (HPF) method is detailed in Section 3. The proposed HPF can improve the effi- ciency of blind flooding by retaining the advantages of statistics-based search mechanisms and by alleviating
  • 2. the partial coverage problem. Section 4 describes our simulation method and the performance metrics. Per- formance evaluation of our proposed HPF method against other search methods is described in Section 5. Section 6 concludes the paper. 2 Search Mechanisms In unstructured P2P systems, the placement of ob- jects is loosely controlled and each peer has no hint where the intended objects are stored. Without having the global knowledge of the dynamic overlay network and the locations of target peers, a source peer has to send a query message to explore as many peers as pos- sible in the overlay network. A well-designed search mechanism should seek to optimize both efficiency and Quality of Service (QoS). Efficiency focuses on better utilizing resources, such as bandwidth and processing power, while QoS focuses on user-perceived qualities, such as number of returned results and response time. In unstructured P2P systems, the QoS of a search mechanism generally depends on the number of peers being explored (queried), response time, and traffic overhead. If more peers can be queried by a certain query, it is more likely that the requested object can be found. In order to avoid having query messages flowing around the network forever, each query message has a TTL (time-to-live: the number of times a query will be forwarded) field. A TTL value is set to limit the search depth of a query. Each time a peer receives a query, the TTL value is decremented by one. The peer will stop relaying the query if TTL becomes zero. A query mes- sage will also be dropped if the query message has vis- ited the peer before. Note that the query messages are application-level messages in an overlay network. In statistics-based search mechanisms, a peer selects a subset of its neighbors to relay the query based on some statistics information of some metrics and heuris- tic algorithms. Based on the number of selected logical query neighbors and the criteria in selecting logical query neighbors, the statistics-based search algorithms in unstructured P2P systems can be roughly classified into two types: uniformed selection of relay neighbors and weighted selection of relay neighbors. 2.1 Uniformed Selection of Relay Neighbors In this approach, all logical neighbors are equally treated when selected to relay the query message. Blind flooding. Blind flooding mechanism relays the query message to all its logical neighbors, except the incoming peer. This mechanism is also referred as breadth-first search (BFS) and is used among peers in Gnutella or among supernodes in KaZaA. For each query, each node records the neighbors which relay the query to it. Thereby on each link, at most two query messages can be sent across it. For an overlay network with m peers and average n neighbors per peer, the total traffic caused by a query is mn if the value of TTL is no less than the diameter of the overlay network. Note that in a typical P2P system, the value of m (more than mil- lions) is much greater than n (less than tens) [13]. In this approach, the source peer can reach its target peer (object) through a shortest path. However, the overhead of blind flooding is very large since flooding generates large amount of unnecessary traffic, wasting bandwidth and processing resource. The simplicity of blind flood- ing makes it very popular in practice. Depth-first search (DFS). Instead of sending que- ries to all the neighbors, a peer just randomly selects a single neighbor to relay the query message when the TTL value is not zero and waits for the response. This search mechanism is referred to as depth-first search (DFS) and is used in Freenet [6]. DFS can terminate timely when the required object has been found, thus avoiding sending out too many unnecessary queries. In DFS, the value of TTL should be set sufficiently large to increase the probability of locating the object. The maximum number of peers that a query message will visit is TTL. Thus, setting a proper TTL value is a key issue to determine the search quality. The response time could be unbearably large due to the nature of its sequential search process. Because of the random selec- tion of relay neighbors, it is possible that an object can hardly be found. K-walker. In k-walker query algorithm proposed in [10], a query is sent to k different walkers (relay neighbors) from the source peer. For a peer in each walker, it just randomly selects one neighbor to relay the query. For each walker, the query processing is done sequentially. For k walkers with up to TTL steps, each query can reach up to k�TTL peers in the P2P network. We can view k-walker search mechanism as a multiple of DFS. It has been shown that k-walker mechanism creates less traffic than that of BFS and provides shorter response time than that of DFS. How- ever, k-walker suffers limited query coverage range due to the randomness nature in selecting query neighbors. 2.2 Weighted Selection of Relay Neighbors Instead of randomly selecting relay neighbors, some mechanisms have been proposed to select relay neighbors more objectively so that neighbors who are most likely to return the requested results are selected. Some statistics information is collected based on some metrics when selecting relay neighbors. Possible met- rics include delay of the link to the corresponding neighbor, the processing time of the neighbor, the com- puting power, the cost (if possible), the amount of shar- ing data, and the number of neighbors, etc.
  • 3. Directed BFS (DBFS). Each peer maintains statistic information based on some metrics, such as the number of results received from neighbors from previous que- ries or the latency of the connection with that neighbor. A peer selects a subset of the neighbors to send its query based on some heuristics, such as selecting the neighbors that have returned the largest number of re- sults from previous queries or selecting the neighbors that have the smaller latency. Routing indices (RI). The concept of routing indices (RI) was proposed in [3]. Each peer keeps a local RI that is a detailed summary of indices, such as the num- ber of files on different topics of interests along each path. When a peer receives a query, it forwards the query to the neighbor that has the largest number of files under a particular topic, rather than selecting relay neighbors at random or flooding to all neighbors. Some weighted-selection search mechanisms have demonstrated performance improvement compared with uniformed-selection search mechanisms. However, weighted-selection search mechanisms have the partial coverage problem to be illustrated in Section 2.4. 2.3 Other Approaches In addition to the aforementioned search policies, there are other techniques that may be used to improve search performance. For example, a peer can cache query responses in hoping that subsequent queries can be satisfied quickly by the cached indices or responses [14, 16, 17]. Peers can also be clustered based on dif- ferent criteria, such as similar interests [14], location in- formation [9], and associative rules [4]. Our proposed statistics-based technique can be used to complement these techniques. 2.4 Partial Coverage Problem Statistics-based search algorithms indeed can reduce network traffic. For example, compared with blind flooding, DBFS can reduce the aggregate processing and bandwidth cost to about 28% and 38%, respectively with 40% increase in the response time [17]. However, our study will show that statistics-based search mecha- nisms may leave a large percentage of the peers un- reachable no matter how large the TTL value is set. We call this phenomena partial coverage problem. This problem is illustrated in Fig.1(a). The number by an edge is the latency between two logical nodes and the number in each node is the number of shared files on that peer. Suppose the size of selected neighbor subset is one and the metric used to select the neighbor is based on the number of shared files. We consider the scenario when the query source is A who has four neighbors (B, C, D, E). It will only send its query to C since C has the largest number of shared files (170). Similarly, C selects D who has the largest number of shared files in all C’s neighbors (B, D, F, G) to relay A’s query. Then D selects A in the same way, which leads to a loop query path: A�C�D�A. Thus, only three nodes are queried in the whole query process while all other nodes are invisible from the query source A. If we change the metric to be the smallest la- tency, the problem still exists because another loop is formed from source A, A�C�B�A. It is very possible that the query cannot be satisfied in the loop. This prob- lem can be less serious when the size of the query sub- set increases, which will be discussed in Section 3. � ��� � �� � � � ��� � ��� � �� � �� �� �� �� �� �� �� �� �� �� �� � � ��� � � �� � �� � �� � ��� � �� � ��� � �� � �� �� � �� ��� �� �� �� �� � � �� � (a) Query path loops (b) Non-optimal query path Figure 1. The partial coverage problem Many statistics-based search approaches use only one metric to collect statistics information to select re- lay neighbors, which does not always lead to an optimal search path. Figure 1(b) shows an example in which A is still the source node. When the search metric is the volume of shared data, the query path would be A�D�E along which the query will check 250 files in 200 unit of time. But obviously if the query path is A�C�G�F�H, the query can check 500 files in 20 units of time. The first path selected using one search metric is not as good as the second one. 3 Hybrid Periodical Flooding In order to effectively reduce the traffic incurred by flooding-based search and alleviate the partial coverage problem, we propose Hybrid Periodical Flooding (HPF). Before discuss HPF, we first define Periodical Flooding. 3.1 Periodical flooding (PF) We notice that in all the existing statistics-based search techniques, the number of relay neighbors, h, does not change at all peers along the query path. In the case of blind flooding, the phenomenon exhibits traffic explosion. The concept of periodical flooding tries to control the number of relay neighbors based on the TTL value along the query path. More specifically, given a peer with n logical neighbors and the current value of TTL, the number of relay neighbors, h, is defined by the following function h=f(n,TTL). Thus, in blind flooding (BFS), we have h=fBFS(n,TTL)=n.. In DFS, we have h=fDFS(n,TTL)=1. The function h=f(n,TTL) can be viewed as a periodi- cal function that changes as TTL changes. We call a
  • 4. search mechanism using a periodical function as peri- odic flooding (PF), in which the query mechanism is divided into several phases that are periodically re- peated. We call the number of different repeated phases as a cycle, C. In all existing statistics-based search tech- niques, they all have a cycle of C=1, which are special cases of PF. We can ask the following questions in or- der to design an efficient search mechanism. In what conditions does a search mechanism with C=1 behave better than a search mechanism with C>1? What is the optimal value of C in terms of a desired performance metric under different underlying physical network to- pologies? For a given C, what is the optimal number of relay neighbors? One example of PF functions with C=2 is shown below: � � � � � � � �� � �� � �� � �� � � evenisTTLifn oddisTTLifn TTLnf , 3 1 , 2 1 ),( �� � �� � � � � � � � � � � �� � �� � � � � � � � � � � ������� �������� (a) BFS (b) PF Figure 2. Comparison between BFS and a PF We compare BFS and the example PF in Fig. 2. Sup- pose peer O initiates a query. Blind flooding (BFS) is employed in Fig. 2(a) where the query is sent or for- warded 36 times to reach all the nodes. We use thin connections to represent the links on which the query traverses once and thick connections to represent the links on which the query traverses twice. We have ex- plained that for each query, each peer records the neighbors, which forward the query to it. Thereby on each link, at most two query messages can be sent across it. When a link is traversed twice, the unneces- sary traffic is incurred. For example, one of the mes- sages from A to B and from B to A is unnecessary. These redundant messages are shown in Fig. 2(a) using dotted arrows. Figure 2(b) illustrates the query process of the ex- ample PF. Peer O has 4 neighbors and has TTL=7. We randomly select relay neighbors. Peer O will select 2 nodes (that is n/2=2 since TTL=7 that is odd), peers A and C, as relay neighbors. Peer A has 5 neighbors. It will select 2 neighbors (G and I) to relay the query initi- ated from peer O since TTL=6 and h=�n/3�=2. Simi- larly, peer C relays the query to peer B and N (TTL=6 and h=�n/3�=2). Although the redundancy problem still exists in PF (such as the traffics from B to J and from I to J), it is significantly reduced compared with that of BFS. Table 1. PF and Blind Flooding TTL Query Msg New Peers Msg Per Peer 7 4 4 1.00 6 17 8 2.12 BFS 5 15 2 7.50 7 2 2 1.00 6 4 4 1.00 PF 5 9 8 1.12 Table 1 compares the redundancy degree of both PF and BFS. It presents the query messages relayed to new peers. For example, in BFS, peers with TTL=5 relay the query to 15 peers, but only 2 of the 15 peers receive the query first time. In PF, peers with TTL=5 relay the query to 9 peer of which 8 are first time receivers. That means for peers with TTL=5, BFS sends 7.5 queries to one new queried peer in average, while PF only sends 1.12 queries to one new queried peer in average. An ef- ficient mechanism should query more peers using less messages. Thus PF is much more efficient than BFS in terms of traffic volume. 3.2 Hybrid Periodical Flooding HPF Overview After determining the number of relay neighbors (h), a peer decides which h nodes should be selected. A simple approach called Random Periodical Flooding (RPF) selects h relay neighbors at random. Selecting re- lay neighbors more objectively may result in better per- formance. For example, we may use the shared data volume as a metric to select query neighbors if we find that peers with more shared data are more likely to sat- isfy queries. By selecting the neighbors with larger number of shard data, a query is more likely to succeed in less number of hops than that of random selection. We may also use the latency between the peer and its neighbors as a metric to select neighbors. In this case, for a given TTL value, a query will experience a shorter delay. If we consider multiple metrics in relay neighbor selection, the search mechanism is expected to have better performance. This motivates us to propose Hy- brid Periodical Flooding (HPF) in which the number of relay neighbors can be changed periodically based on a periodical function and the relay neighbors are selected based on multiple metrics in a hybrid way. HPF differentiates with RPF in that RPF selects re- lay neighbors randomly, and differentiates with DBFS in that DBFS only uses one metric to select relay neighbors. HPF selects neighbors based on multiple metrics and provides flexibility to justify different pa- rameters to improve overall performance. Let h denote the expected number of relay neighbors, which is given
  • 5. by h = h1 + h2 + … + ht, where t is the number of met- rics used in relay neighbor selection and hi is the num- ber of relay neighbors selected by metric i. Metrics There are many metrics that may be used to select re- lay neighbors, such as communication cost, bandwidth, number of returned results from the neighbor, average number of hops from the neighbor to peers who re- sponded the previous queries, and so on. These metrics may have different weights for a system with different query access patterns or different performance require- ments. For example, we may give higher weights to some metrics that are more sensitive to the performance in a specific system. We have �� � t i iw 1 1 , where iw is the weight assigned to metric i ( ti ��1 ). To alleviate the partial coverage problem, we select relay neighbors in a hybrid way. We select hi neighbors using metric i, where hi is determined by � �ii whh �� . Let Si denote the set of neighbors selected based on the metric i. The complete set of relay neighbors is i t i SS 1� � � , where |S| iih � . Note that a neighbor may be selected by more than one metric. Thus, the actual number of relay neighbors selected may be less than h. Termination of Search Queries A query process is terminated when a pre-set TTL value has been decreased to zero. Choosing an appro- priate TTL value is very difficult. A large TTL may cause higher traffic volume, while a small TTL may not respond with enough number of query results. Further- more there are no mutual feedbacks between the source peer and the peers who forward or respond the query. Thus it is hard for peers to know when to stop forward- ing the query before the TTL value is reduced to zero. Iterative Deepening [17] made an effort to address this problem in some degree. In Iterative Deepening, a policy P is used to control the search mechanism, which provides a sequence of TTLs so that a query is flooded from a very small TTL, and if necessary, to a gradually enlarged scope. For example, one policy can be P={a, b, c}, where P has three iterations. A query starts to be flooded with TTL=a. If the query cannot be satisfied, it will be flooded with TTL=b-a from all peers that are a hops away from the source peer. Similarly if the query still cannot be satisfied, it will be flooded with TTL=c-b from all peers that are b hops away from the source peer. In this policy, c is the maximal length of a query path. Iterative Deepening is a good mechanism in the sense that it alleviates the process time of middle nodes be- tween iterations. In HPF, we use this policy to terminate the success- ful queries without incurring too much unnecessary traffic. Since the combination is quite straightforward and the performance of Iterative Deepening policy has been evaluated in [17], this policy will not be re- evaluated in this paper. 4 Simulation Methodology We use simulation to evaluate the performance of RPF and HPF and analyze the effects of the parameters. 4.1 Topology Generation Two types of topologies, physical topology and logi- cal topology, have to be generated in our simulation. The physical topology should represent the real topol- ogy with Internet characteristics. The logical topology represents the overlay P2P topology built on top of the physical topology. All P2P nodes are in the node subset of the physical topology. The communication cost be- tween two logical neighbors is calculated based on the physical shortest path between this pair of nodes. To simulate the performance of different search mecha- nisms in a more realistic environment, the two topolo- gies must accurately reflect the topological properties of real networks in each layer. Previous studies have shown that both large scale Internet physical topologies [15] and P2P overlay to- pologies follow small world and power law properties. Power law describes the node degree while small world describes characteristics of path length and clustering coefficient [2]. Studies in [12] found that the topologies generated using the AS Model have the properties of small world and power law. BRITE [1] is a topology generation tool that provides the option to generate to- pologies based on the AS Model. Using BRITE, we generate 10 physical topologies each with 10,000 nodes. The logical topologies are generated with the number of peers ranging from 1,000 to 5,000. The average number of edges of each node is ranging from 6 to 20. 4.2 Simulation Setup The total network traffic incurred by queries and av- erage response time of all queries are two major metrics that we use to evaluate the efficiency of a search mechanism. High traffic volume will limit system scal- ability and long response time is intolerable for users. Network administrators care more about how much network bandwidth consumed by a P2P system, while users care more about the response time of queries, which is viewed as a part of service quality of the sys- tem.
  • 6. 0 5 10 15 20 25 30 35 40 45 50 0 1% 2% 3% 4% 5% 6% 7% 8% Coverage Size NodesDistribution(%) 1,000-node overlay network 10,000-node physical network 400 410 420 430 440 450 460 470 480 490 500 0 1% 2% 3% 4% 5% 6% 7% 8% 9% NodesDistribution(%) Coverage Size 1,000-node overlay network 10,000-node physical network 0 10 20 30 40 50 60 70 80 90 100 0 0.5% 1.0% 1.5% 2.0% 2.5% 3.0% 3.5% 4.0% NodesDistribution(%) Coverage Size 1,000-node overlay network 10,000-node physical network Figure 3. Node distribution vs. cover- age size. (h=1, metric 2) Figure 4. Node distribution vs. cover- age size. (h=2, metric 2) Figure 5. Node distribution vs. cover- age size. (h=1, metric 1) In our simulation, we consider two metrics with the same weight to select relay neighbors in HPF. In prac- tice, more metrics could be used for neighbor selection. The two metrics are the communication cost (metric 1) that is the distance between a peer and its neighbor and the shared number of files (metric 2) on each node. Based on the first metric, a peer will select the neighbors with the less communication costs. Based on the second metric, a peer will select the neighbors with the larger amount of shared data. For each given search criterion, we distribute 100 files satisfying the search on the peers in a generated P2P topology. That means there are totally 100 possible results for a specific query in the whole P2P network. The distribution of the 100 files on the network is ran- dom. For each peer, we generate a number within 1 to 1000 as the number of shared files in this peer. Based on the second metric in selecting relay neighbors, a neighbor with more shared files is more likely to return a response than a neighbor with less shared files. 5 Performance Evaluation In this section, we present the simulation results to show the effectiveness of HPF compared with DBFS and BFS. 5.1 Partial Coverage Problem Based on [3, 17], statistics-based search mechanisms are more efficient and incur less traffic to the Internet compared with blind flooding. However, statistics- based search mechanisms have partial coverage prob- lem as we discussed in Section 2.4. We quantitatively illustrate the partial coverage problem in this section. We first illustrate the case in which only one relay neighbor is selected to send/forward a query (h=1) based on the number of shared files in neighbors. We set TTL as infinity. Figure 3 shows the node distribution versus the number of peers being queried, which is de- fined as coverage size. For example, queries initiated from 8% of peers can only reach 10 other peers. Most of peers can only push their queries to 10 to 30 other peers. This means that loops are formed and only a very small number of peers can be reached for any queries. Note that the overlay network has 1000 nodes and the physical network has 10,000 nodes. Figure 4 illustrates the node distribution versus the coverage size, where h=2 and TTL=infinity. The coverage size is about 400 peers in average, which is still a small number in a P2P network. Figure 5 shows node distribution versus coverage size when we use network latency as the metric to se- lect relay neighbors. Again, we see the partial coverage problem. The partial coverage problem will disappear when h=n, which is the case of blind flooding. We did the same group of simulations on different topologies using different metrics. The results are quite consistent. Figure 6 shows the percentage of covered peers to total peers versus the number of relay neighbors (h=1, 2, n/5, n/4, n/3, n/2, and Sqrt(n)). The percentage of coverage is larger for a larger h. A larger h means a smaller chance for all reached peers to form a loop. One Two 1/5 1/4 1/3 Half Sqrt 0 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Number of Relay Neighbors PercentageofCoveredNodes 1.65% 42.4% 56.3% 63.0% 77.5% 87.5% 77.6% Figure 6. Percentage of coverage vs. the number of relay neighbors 5.2 Performance of Random PF We have evaluated network traffic and average re- sponse time of RPF that selects relay neighbors at ran- dom. We can use many different periodical flooding functions to determine the number of relay neighbors. These functions should not be over complicated. We have tried tens of periodical flooding functions with dif- ferent C.
  • 7. 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 80 90 100 Requested Number of Response Results NormalizedQueryCost BFS RPF (1) RPF (2) RPF (3) 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 80 90 100 Requested Number of Response Results NormalizedResponseTime BFS RPF (1) RPF (2) RPF (3) DFS 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 80 90 100 Requested Number of Response Results NormalizedQueryCost BFS RPF DBFS HPF Figure 7. Normalized traffic of RPF Figure 8. Normalized response time of RPF Figure 9. Normalized traffic com- parison Figures 7 and 8 show the normalized network traffic cost and normalized average response time versus the required number of response results. The traffic and av- erage response time always perform in opposite way. If a search mechanism causes low traffic, it will suffer from high response time and vice versa. RPF is de- signed to provide an opportunity to have a tradeoff be- tween total traffic and average response time, thus ob- taining a better overall search performance. We may expect a search mechanism to reduce a large amount of traffic by increasing a little more response time or vice versa. How to quantitatively measure the overall per- formance based on the tradeoff is an issue. It’s hard to find the best search mechanism. We de- fine p to measure the overall performance, where timetrafficp RC �� �� , traffic and time are nor- malized value of total network traffic and average re- sponse time, �C and �R are the weight parameters for network traffic and response time, and �C + �R = 1. We seek an asymptotically periodical flooding function fa(n,TTL) such that p can be minimal or close to mini- mal. If a system emphasizes more on low network traf- fic, we can set �C > �R; otherwise, we can set �C < �R for a system emphasizing more on quick response time. Based on different topologies with different number of average connections, and different values of �C and �R, the functions of fa(n,TTL) may be derived differently. In our simulation of HPF, the average number of edge connections is 10. We choose �C = 0.6 and �R = 0.4. Thus, the corresponding period function is derived as: � � � � � � � �� � �� � �� � �� � � evenisTTLifn oddisTTLifn TTLnf , 4 1 , 2 1 ),( 5.3 Effectiveness of HPF HPF selects relay neighbors based on multiple met- rics in a hybrid way. We use communication cost and the volume of shared data as two metrics to select relay neighbors. Based on the simulation over 10,000 queries, Figure 9 shows the normalized network traffic versus the re- quired number of response results of four different search mechanisms: BFS, RPF, DBFS and HPF. DBFS reduces the network traffic by 30~50% compared with BFS. HPF outperforms DBFS by up to 20%. Figure 10 compares the normalized response time of four different search mechanisms over 10,000 queries versus the re- quired number of response results. HPF performs the best compared with RPF and DBFS, but still worse than BFS. DBFS selects relay neighbors who have the larg- est volume of shared files. Each query may get more re- sults by reaching fewer peers. HPF needs to query more peers to obtain the same amount of results than DBFS but much less than BFS and RPF. That is because we use multiple metrics instead of a single metric used in DBFS, expecting to obtain better overall performance, which has been shown in Figs. 9 and 10. 5.4 Alleviating the Partial Coverage Problem HPF can effectively address the partial coverage problem discussed in Section 2.4. Figure 11 shows the percentage of queried peers as TTL increases. BFS can quickly cover 100% peers, while DBFS can only cover up to 77% peers in our simulation because of the partial coverage problem. DBFS still covers only around 77% when the value of TTL is set to infinity in our simula- tion. However, HPF and RPF can cover more than 96% peers as TTL is increased to 10. Figure 12 compares the peer coverage size of DBFS and HPF. In DBFS, most nodes can cover 760-780 peers out of 1,000 nodes. The coverage size is increased to 950-970 in HPF.
  • 8. 0 5 10 15 20 25 30 35 40 0 10 20 30 40 50 60 70 80 90 100 Requested Number of Response Results NormallizedResponseTime BFS RPF DBFS HPF 0 2 4 6 8 10 12 14 0 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% TTL PercentageofCoveredNodes BFS HPF DBFS RPF 600 650 700 750 800 850 900 950 1000 0 2% 4% 6% 8% 10% 12% 14% 16% 18% Coverage Size NodesDistribution DBFS HPF Figure 10. Normalized response time comparison Figure 11. Coverage percentage comparison Figure 12. Partial coverage comparison 6 Conclusion In this paper, we have proposed an efficient and adaptive search mechanism, Hybrid Periodical Flooding. HPF improves the efficiency of blind flooding by re- taining the advantages of statistics-based search mecha- nisms and by alleviating the partial coverage problem. We summarize our contributions as follows: � Analyze the current search mechanisms used and proposed in unstructured P2P networks. � Qualitatively and quantitatively analyze the partial coverage problem caused by statistics-based search mechanisms, such as DBFS. � Propose to use a periodical flooding function to de- fine the number of relay neighbors, which can be adaptively changed. This is the first technique used in HPF. � Propose to use multiple metrics to select relay neighbors to obtain better overall performance or adaptively meet different performance requirements, which is the second technique used in HPF. We have shown the performance of HPF using two metrics to select relay neighbors. HPF provides the flexibility to use more metrics and allows the applica- tion to define multiple metrics and give them different weights, thereby the algorithm is more flexible in prac- tice to meet different performance requirements. References [1] BRITE, http://www.cs.bu.edu/brite/. [2] T. Bu and D. Towsley, On distinguishing between Inter- net power law topology generators, In Proceedings of IEEE INFOCOM'02 Conference, 2002. [3] A. Crespo and H. Garcia-Molina, Routing indices for peer-to-peer systems, In Proceedings of 22nd Interna- tional Conference on Distributed Computing Systems, 2002. [4] E.Cohen, A.Fiat, and H.Kaplan, Associative search in peer to peer networks: harnessing latent semantics, In Proceedings of the IEEE INFOCOM'03, 2003. [5] Fasttrack, http://www.fasttrack.nu/. [6] Freenet, http://freenet.sourceforge.net. [7] Gnutella, http://gnutella.wego.com/. [8] KaZaA, http://www.kazaa.com. [9] B. Krishnamurthy and J. Wang, Automated traffic classi- fication for application-specific peering, In Proceedings of ACM SIGCOMM Internet Measurement Workshop, November 2002. [10] Q. Lv, et al., Search and replication in unstructured peer- to-peer networks, In Proceedings of the 16th ACM Inter- national Conference on Supercomputing, 2002. [11] Ritter, Why Gnutella can't scale. No, really. http://www.tch.org/gnutella.html. [12] S. Saroiu, P. Gummadi, and S. Gribble, A measurement study of peer-to-peer file sharing systems, In Proceedings of Multimedia Computing and Networking (MMCN), 2002. [13] S. Sen and J. Wang, Analyzing peer-to-peer traffic across large networks, In Proceedings of ACM SIG- COMM Internet Measurement Workshop, 2002. [14] K. Sripanidkulchai, B. Maggs, and H. Zhang, Efficient content location using interest-based locality in peer-to- peer systems, In Proceedings of INFOCOM'03, 2003. [15] H. Tangmunarunkit, et al., Network topology generators: degree-based vs. structural, In Proceedings of In Pro- ceedings of SIGCOMM'02, 2002. [16] B. Yang and H. Garcia-Molina, Designing a super-peer network, In Proceedings of the 19th International Con- ference on Data Engineering (ICDE), March 2003. [17] B. Yang and H. Garcia-Molina, Efficient search in peer- to-peer networks, In Proceedings of ICDCS'02, 2002.