P24120125

  • 158 views
Uploaded on

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

More in: Technology
  • Full Name Full Name Comment goes here.
    Are you sure you want to
    Your message goes here
    Be the first to comment
    Be the first to like this
No Downloads

Views

Total Views
158
On Slideshare
0
From Embeds
0
Number of Embeds
0

Actions

Shares
Downloads
0
Comments
0
Likes
0

Embeds 0

No embeds

Report content

Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
    No notes for slide

Transcript

  • 1. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-125 Data Caching Placement based on information density in wireless ad hoc network C.Srinivas, Samreen KhanAbstract network is generally resource constrained in terms of Data caching strategy for ad hoc networks whose channel bandwidth or battery power in the nodes.nodes exchange information items in a peer-to-peer Caching helps in reducing communication, this results infashion. Data caching is a fully distributed scheme where savings in bandwidth, as well as battery energy. Theeach node, upon receiving requested information, problem of cache placement is particularly challengingdetermines the cache drop time of the information or when each network node has a limited memory to cachewhich content to replace to make room for the newly data items.arrived information. These decisions are madedepending on the perceived ―presence‖ of the content in In this paper, our focus is on developingthe nodes proximity, whose estimation does not cause efficient caching techniques in ad hoc networks withany additional overhead to the information sharing memory limitations. Research into data storage, access,system. and dissemination techniques in ad hoc networks is not new. In particular, these mechanisms have beenWe devise a strategy where nodes, independent of each investigated in connection with sensor networking peer-other, decide whether to cache some content and for how to-peer networks mesh networks world wide Web andlong. In the case of small-sized caches, we aim to design even more general ad hoc networks. However, thea content replacement strategy that allows nodes to presented approaches have so far been somewhat ―adsuccessfully store newly received information while hoc‖ and empirically based, without any strongmaintaining the good performance of the content analytical foundation. In contrast, the theory literaturedistribution system. Under both conditions, each node abounds in analytical studies into the optimalitytakes decisions according to its perception of what properties of caching and replica allocation problems.nearby users may store in their caches and with the aim However, distributed implementations of theseof differentiating its own cache content from the other techniques and their performances in complex networknodes’. settings have not been investigated. It is even unclear whether these techniques are amenable to efficient The result is the creation of content diversity within distributed implementations.the nodes neighbourhood so that a requesting user likelyfinds the desired information nearby. We simulate our Our goal in this paper is to develop an approachcaching algorithms in different ad hoc network scenarios that is both analytically tractable with a provableand compare them with other caching schemes, showing performance bound in a centralized setting and is alsothat our solution succeeds in creating the desired content amenable to a natural distributed implementation. In ourdiversity, thus leading to a resource-efficient information network model, there are multiple data items; each dataaccess. item has a server, and a set of clients that wish to access the data item at a given frequency. Each node carefullyKey terms: Data Caching, Distributed Caching chooses data items to cache in its limited memory toAlgorithm, Ad hoc network minimize the overall access cost. Essentially, in this article, we develop efficient strategies to select dataI. INTRODUCTION items to cache at each node. In particular, we develop Ad hoc networks are multi hop wireless two algorithms—a centralized approximation algorithm,networks of small computing devices with wireless which delivers a 4-approximation (2-approximation forinterfaces. The computing devices could be conventional uniform size data items) solution, and a localizedcomputers (for example, PDA, laptop, or PC) or distributed algorithm, which is based on thebackbone routing platforms or even embedded approximation algorithm and can handle mobility ofprocessors such as sensor nodes. The problem of optimal nodes and dynamic traffic conditions. Using simulations,placement of caches to reduce overall cost of accessing we show that the distributed algorithm performs verydata is motivated by the following two defining close to the approximation algorithm. Finally, we showcharacteristics of ad hoc networks. First, the ad hoc through extensive experiments on ns2 that our proposednetworks are multi hop networks without a central base distributed algorithm performs much better than a priorstation. Thus, remote access of information typically approach over a broad range of parameter values. Ours isoccurs via multi hop routing, which can greatly benefit the first work to present a distributed implementationfrom caching to reduce access latency. Second, the based on an approximation algorithm for the general 120 | P a g e
  • 2. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-125problem of cache placement of multiple data items under how long, resulting in a distributed scheme that does notmemory constraint. require additional control messages. Data caching strategy for ad hoc networks The Hamlet approach applies to the following cases.whose nodes exchange information items in a peer-to- • Information presence estimation. In this case wepeer fashion. Data caching is a fully distributed scheme define the reach range of a generic node that can receivewhere each node, upon receiving requested information, a query generated by node n itself. As an example, in andetermines the cache drop time of the information or ideal setting in which all nodes have the same radiowhich content to replace to make room for the newly range, the reach range is given by the product of TTLarrived information. These decisions are made and the node radio range. Next we denote by f thedepending on the perceived ―presence‖ of the content in frequency at which every node estimate the presence ofthe nodes proximity, whose estimation does not cause each information item with in the reach range , and weany additional overhead to the information sharing define as 1/f the duration of each estimation step (alsosystem. called time step hereafter). The generic node n uses the information captured We devise a strategy where nodes, independent with in its reach range, during the estimation step j, toof each other, decide whether to cache some content and compute the following quantities:1) a provider counterfor how long. In the case of small-sized caches, we aim by using application-layer data and 2) a transit counterto design a content replacement strategy that allows by using data that are collected through channelnodes to successfully store newly received information overhearing in a cross-layer fashion. These counters arewhile maintaining the good performance of the content defined as follows:distribution system. Under both conditions, each node • Provider counter dic(n, j). This quantity accounts fortakes decisions according to its perception of what the presence of new copies of information i’s chunks c,nearby users may store in their caches and with the aim delivered by n to querying nodes within its range duringof differentiating its own cache content from the other step j. Node n updates this quantity every time it acts asnodes’. The result is the creation of content diversity a provider node.( e.g., like node P in the upper plot ofwithin the nodes neighbourhood so that a requesting user figure 2)likely finds the desired information nearby. We simulateour caching algorithms in different ad hoc networkscenarios and compare them with other caching schemes,showing that our solution succeeds in creating thedesired content diversity, thus leading to a resource-efficient information access.II. SYSTEM OUTLINE OVERVIEW: Hamlet is fully distributed caching wireless adhoc networks whose nodes exchange information item ina peer to peer fashion. In particular we address a mobilead hoc network whose nodes might be resource-constrained devices, pedestrian users, or vehicles on cityroads. Each node runs an application to request andpossibly cache desired information items. Nodes in the • Large-sized caches. In this case, nodes can potentiallynetwork retrieve information items from other users that store a large portion (i.e., up to 50%) of the availabletemporarily cache (part of ) the requested items or from information items. Reduced memory usage is a desirableone or more gate way nodes, which can store content or (if not required) condition, because the same memoryquickly fetch it from the internet . may be shared by different services and applications that We propose, called Hamlet, aims at creating content run at nodes. In such a scenario, a caching decisiondiversity within the node neighbourhood so that users consists of computing for how long a given contentlikely find a copy of the different information items should be stored by a node that has previously requestednearby (Regardless of the content popularity level) and it, with the goal of minimizing the memory usageavoid flooding the network with query messages. without affecting the overall information retrievalAlthough a similar concept has been put forward in the performance;novelty in our proposal resides in the probabilisticestimate, run by each node, of the information presence • Small-sized caches. In this case, nodes have a dedicated(i.e., of the cached content) in the node proximity. The but limited amount of memory where to store a smallestimate is performed in a cross-layer fashion by percentage (i.e., up to 10%) of the data that they retrieve.overhearing content query and information reply The caching decision translates into a cache replacementmessages due to the broadcast nature of the wireless strategy that selects the information items to be droppedchannel. By leveraging such a local estimate, nodes among the information items just received and theautonomously decide what information to keep and for information items that already fill up the dedicated 121 | P a g e
  • 3. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-125memory. We evaluate the performance of Hamlet in clarity of presentation, we assume uniform-sizedifferent mobile network scenarios, where nodes (occupying unit memory) data items for now. Ourcommunicate through ad hoc connectivity. The results techniques easily generalize to non-uniform size datashow that our solution ensures a high query resolution items, as discussed later. Each node i has a memoryratio while maintaining the traffic load very low, even capacity of mi units. We use aij to denote the accessfor scarcely popular content, and consistently along frequency with which a node i requests the data item Dj ,different network connectivity and mobility scenarios. and dil to denote the weighted distance between two network nodes i and l. The cache placement problem isIII. System Analysis to select a set of sets M = {M1, M2. . . ,Mp}, where Mj is a set of networknodes that store a copy of Dj , to minimize In this we are going to deal with the caching the total access costplacement problems and algorithm. Those are as follows: 1. SELF-ORGANIZING under the memory capacity constraint that 2. SELF-ADDRESSING (MANET) 3. INTEGRATED CACHE-ROUTING which means each network node i appears in at most mi sets of M. The cache placement problem is known to be 4. LOCALIZED CACHING POLICY NP-hard [3]. EXAMPLE 1: Figure 1 illustrates the above described 5. DISTRIBUTED CACHING ALGORITHM cache placement problem in a small ad hoc network. In Figure 2, each graph edge has a unit weight. All the 1. SELF-ORGANIZING nodes have the same memory capacity of 2 pages, and Multiple Data’s deals with the three algorithms the size of each data item is 1 memory page. Each of thefor cache placement of multiple data items in ad hoc nodes 1, 2, 3, 4, and 12 have one distinct data item to benetworks. In the first approach, each node caches the served (as shown in the parenthesis with their nodeitems most frequently accessed by itself; the second numbers). Each of the client nodes (9, 10, 11, 13, and 14)approach eliminates replications among neighboring accesses each of the data items D1;D2, and D3 with unitnodes introduced by the first approach; the third access frequency. Figure 2 shows that the nodes 5, 6, 7,approach require one or more ―central‖ nodes to gather 8 have cached one or more data items, and also showsneighbourhood information and determine caching the cache contents in those nodes. As indicated by theplacements. The first two approaches are largely bold edges, the clients use the nearest cache node insteadlocalized, and hence, would fare very badly when the of the server to access a data item. The set of cachepercentage of client nodes in the network is low, or the nodes of each data item are: M1 = {7, 8},M2 = {7, 8}, M3access frequencies are uniform. In the third approach, it = {5, 6}. One can observe that total access cost is 20is hard to find stable groups in ad hoc networks because units for the given cache placement.of frequent failures and movements.2. SELF-ADDRESSING (MANET): A multi-hop ad hoc network can be representedas an undirected graph G(V, E) where the set of nodesvertices V the nodes in the network and E are the set ofweighted edges in the graph. Two network nodes thatcan communicate directly with each other are connectedby an edge in the graph. The edge weight may representa link metric such as loss rate, delay, or transmissionpower. For the cache placement problem addressed inthis article, there are multiple data items and each dataitem is served by its server (a network node may act as aserver for more than one data items). Each network nodehas limited memory and can cache multiple data itemssubject to its memory capacity limitation. The objective Fig. 2. Illustrating cache placement problem underof our cache placement problem is to minimize the memoryconstraintoverall access cost. Below, we give a formal definitionof the cache placement problem addressed in this system. Problem Formulation. Given a general ad hocnetwork graph G(V;E) with p data items D1,D2………,Dp,where a data item Dj is served by a server Sj . A networknode may act as a server for multiple data items. For 122 | P a g e
  • 4. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-1253. Integrated Cache Routing In addition, the inaccuracies and staleness of the nearest cache table entries (due to message losses or Integrated cache routing is based on three step arbitrary communication delays) may result inprocedure those are as follows: approximate local benefit values. Finally, in DGA, the  Nearest-caching tables can be used in placement of caches happens simultaneously at all nodes conjunction with any underlying routing in a distributed manner, which is in contrast to the protocol to reach the nearest cache node, as sequential manner in which the caches are selected by long as the distances to other nodes are the CGA. However, DGA is able to cope with maintained by the routing protocol. dynamically changing access frequencies and cache placements. As noted before, any changes in cache  However, note that maintaining cache-routing placements trigger updates in the nearest-cache table, tables instead of nearest-cache tables and which in turn affect the local benefit values. Below is a routing tables doesn’t offer any clear advantage summarized description of the DGA. in terms of number of messages transmissions. Algorithm 1: Distributed Greedy Algorithm (DGA) Setting We could maintain the integrated cache-routing A network graph G(V;E) with p data items.tables in the similar vein as routing tables are maintained Each node i has a memory capacity of mi pages. Let in mobile ad hoc networks. Alternatively, we could have be the benefit threshold.the servers periodically broadcast the latest cache lists. Program of Node iIn our simulations, we adopted the latter strategy, since it BEGINprecludes the need to broadcast Add Cache and Delete When a data item Dj passes byCache messages to some extent. if local memory has available space and (Bij >  )4. LOCALIZED CACHING POLICY then cache Dj  The caching policy of DGA is as follows. Each else if there is a set D of cached data items such node computes benefit of data items based on that(local benefit of D < Bij   and (|D|  |Dj |), ) its ―local traffic‖ observed for a sufficiently then replace D with Dj long time. The local traffic of a node i includes When a data item Dj is added to local cache its own local data requests, non-local data Notify the server of Dj requests to data items cached at i, and the traffic Broadcast an AddCache message with (i ,Dj) that the node i is forwarding to other nodes in When a data item Dj is deleted from local cache the network. Get the cache list Cj from the server of Dj Broadcast a DeleteCache message with  A node decides to cache the most beneficial (in (i’,Dj ,Cj) terms of local benefit per unit size of data item) On receiving an AddCache message data items that can fit in its local memory. (i’,Dj) When the local cache memory of a node is full, if i’ is nearer than current nearest- the following cache replacement policy is used. cache for Dj then update nearest-cache table entry  In particular, a data item is newly cached only if and broadcast the AddCache its local benefit is higher than the benefit message to neighbors threshold, and a data item replaces a set of else send the message to the nearest- cached data items only if the difference in their cache of i local benefits is greater than the benefit On receiving a DeleteCache message threshold. (i0, Dj, Cj ) if i’ is the current nearest-cache for Dj 6. DISTRIBUTED GREEEDY ALGORITHM then update the nearest-case of Dj (DGA) using Cj , and broadcast the Distributed Greedy Algorithm (DGA) is DeleteCache messagealso known as Distributed Caching Algorithm (DCA). else send the message to the nearest- The above components of nearest-cache cache of itable and cache replacement policy are combined to For mobile networks, instead ofyield our Distributed Greedy Algorithm (DGA) for AddCache and DeleteCachecache placement problem. In addition, the server uses the messages, for each data item, itscache list to periodically update the server periodically broadcasts (tocaches in response to changes to the data at the server. the entire network) the latestThe departure of DGA from CGA is primarily in its cache list.inability to gather information about all traffic (access END.frequencies). 123 | P a g e
  • 5. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-125Performance Analysis. Based on the above observation,  We have developed efficient data cachingif we assume that local benefit is reflective of the algorithms to determine near optimal cacheaccurate benefit (i.e., if the local traffic seen by a node i placements to maximize reduction in overallis the only traffic that is affected by caching a data item access cost. Reduction in access cost leads toat node i), then DGA also yields a solution whose benefit communication cost savings and hence, betteris one fourth of the optimal benefit. bandwidth usage and energy savings.Data Expiry and Cache Updates. We incorporate theconcepts of data expiry and cache updates in our overall  However, our simulations over a wide range offramework as follows. For data expiry, we use the network and application parameters show thatconcept of Time-to-Live (TTL) [7], which is the time till the performance of the caching algorithms.which the given copy of the data item is consideredvalid/fresh. The data item or its copy is considered  Presents a distributed implementation based onexpired at the an approximation algorithm for the problem ofend of the TTL time value. We consider two data expiry cache placement of multiple data items undermodels, viz. TTL-per-request and TTL-per-item. In memory constraint. the TTL-per-request data expiry model [7], the server responds to any data item request with the requested  The result is the creation of content diversity data item and an appropriately generated TTL value. within the nodes neighborhood so that a Thus, each copy of the data item in the network is requesting user likely finds the desired associated with an independent TTL value. In the TTL- information nearby. per-item data expiry model, the server associates a TTL value with each data item (rather than each  We simulate our caching algorithms in different request), and all requests for the same data item are ad hoc network scenarios and compare them associated with the same TTL (until the data item with other caching schemes, showing that our expires). Thus, in the TTL-per-item data expiry model, solution succeeds in creating the desired content all the fresh copies of a data item in the network are diversity, thus leading to a resource-efficient associated with the same TTL value. On expiry of the information access. data item, the server generates a new TTL value for the data item. For updating the cached data items, we LIMITATIONS & FUTURE consider two mechanisms. For the case of TTL-per- ENHANCEMENTS : request data expiry model, we use the cache deletion Hamlet is caching self contained and is designed to self update model, where each cache node independently adapt to network environments with different mobility deletes its copy of the expired data item. Such and connectivity features. deletions are handled in the similar way as describe We assume a content distribution system where the before, i.e., by broadcasting a DeleteCache request. In following assumptions hold: the case of TTL-per-item data expiry model, all the copies of a particular data item expire simultaneously. 1) A number I of information items is available to the Thus, we use the server multicast cache update model, users, with each item divided into a number C of chunks; wherein the server multicasts the fresh copy of the data item to all the cache nodes, on expiration of the 2) User nodes can overhear queries for content and data item (at the server). If the cache list is not relative responses within their radio proximity by maintained at the server, then the above update is exploiting the broadcast nature of the wireless medium; implemented using a network wide broadcast andCONCLUSION: 3) User nodes can estimate their distance in hops from  Data caching strategy for ad hoc networks the query source and the responding node due to a hop- whose nodes exchange information items in a count field in the messages. Although Hamlet can work peer-to-peer fashion. Data caching is a fully with any system that satisfies the aforementioned three distributed scheme where each node, upon generic assumptions, for concreteness, we detail the receiving requested information, determines the features of the specific content retrieval system that we cache drop time of the information or which will consider in the remainder of this paper. content to replace for the newly arrived information. REFERENCE & BIBLIOGRAPHY:  We have developed a paradigm of data caching Good Teachers are worth more than thousand techniques to support effective data access in ad books, we have them in Our Department. hoc networks. In particular, we have considered [1] Caching Strategies Based on Information Density memory capacity constraint of the network Estimation in Wireless Ad Hoc NetworksMarco nodes. Fiore, Member, IEEE, Claudio Casetti, Member, IEEE, and Carla-Fabiana Chiasserini, Senior 124 | P a g e
  • 6. C.Srinivas, Samreen Khan / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue 4, July-August 2012, pp.120-125 Member, IEEE IEEE TRANSACTIONS ON dissemination,‖ IEEE Trans. Knowl. Data Eng., VEHICULAR TECHNOLOGY, VOL. 60, NO. 5, vol. 16, no. 1, pp. 125–139, Jan. 2004. JUNE 2011 pg.no.1294 to pg.no.2028 [14] J. Cao, Y. Zhang, G. Cao, and L. Xie, ―Data[2] B.-J. Ko and D. Rubenstein, ―Distributed self- consistency for cooperative caching in mobile stabilizing placement of replicated resources in environments,‖ Computer, vol. 40, no. 4, pp. 60– emerging networks,‖ IEEE/ACM Trans. Netw., 66, Apr. 2007. vol. 13, no. 3, pp. 476–487, Jun. 2005. [15] N. Dimokas, D. Katsaros, and Y. Manolopoulos,[3] ―Benefit-based Data Caching in Ad Hoc ―Cache consistency in wireless multimedia sensor Networks ‖ Bin Tang, Member, IEEE, Himanshu networks,‖ Ad Hoc Netw., vol. 8, no. 2, pp. 214– Gupta, Member, IEEE, and Samir R. Das, 240, Mar. 2010. Member, IEEE [16] M. Fiore, F. Mininni, C. Casetti, and C.-F.[4] G. Cao, L. Yin, and C. R. Das, ―Cooperative Chiasserini, ―To cache or not to cache?‖ in Proc. cache-based data access in ad hoc networks,‖ IEEE INFOCOM, Rio de Janeiro, Brazil, Apr. Computer, vol. 37, no. 2, pp. 32–39, Feb. 2004. 2009, pp. 235–243.[5] C.-Y. Chow, H. V. Leong, and A. T. S. Chan, REFERENCES MADE FROM: ―GroCoca: Group-based peer-to-peer cooperative caching in mobile environment,‖ IEEE J. Sel. 1. Professional Java Network Programming Areas Commun., vol. 25, no. 1, pp. 179–191, Jan. 2. Java Complete Reference 2007. 4. Data Communications and Networking, by Behrouz A Forouzan.[6] T. Hara, ―Cooperative caching by mobile clients 5. Computer Networking: A Top-Down Approach, by in push-based information systems,‖ in Proc. James F. Kurose. CIKM, 2002, pp. 186–193. 6. Operating System Concepts, by Abraham Silberschatz.[7] L. Yin and G. Cao, ―Supporting cooperative C.Srinivas M.Tech Associate Professor , HOD caching in ad hoc networks,‖ IEEE Trans. Mobile IT Department, Sree Visvesvraya Institute of technology Comput., vol. 5, no. 1, pp. 77–89, Jan. 2006. and Sciences affiliated to JNTUH, approved by AICTE New Delhi, Mahabubnagar. His research includes[8] N. Dimokas, D. Katsaros, and Y. Manolopoulos, Information security , Networking ―Cooperative caching in wireless multimedia sensor networks,‖ ACM Mobile Netw. Appl., vol. 13, no. 3/4, pp. 337–356, Aug. 2008. Samreen Khan pursuing M.Tech CSE[9] Y. Du, S. K. S. Gupta, and G. Varsamopoulos, final year in Sree Visvesvaraya ―Improving on-demand data access efficiency in Institute of Technology and Science MANETs with cooperative caching,‖ Ad Hoc affiliated to JNTUH, approved by Netw., vol. 7, no. 3, pp. 579–598, May 2009. AICTE New Delhi. Chowderpally Mahabubnagar.[10] W. Li, E. Chan, and D. Chen, ―Energy-efficient Research area includes Java cache replacement policies for cooperative computer network, information security . caching in mobile ad hoc network,‖ in Proc. IEEE WCNC, Kowloon, Hong Kong, Mar. 2007, pp. 3347–3352.[11] M. K. Denko and J. Tian, ―Cross-layer design for cooperative caching in mobile ad hoc networks,‖ in Proc. IEEE CCNC, Las Vegas, NV, Jan. 2008, pp. 375–380.[12] H. Chen, Y. Xiao, and X. Shen, ―Update-based cache replacement policies in wireless data access,‖ in Proc. BroadNets, Boston, MA, Oct. 2005, pp. 797–804.[14] J. Xu, Q. Hu, W.-C. Lee, and D. L. Lee, ―Performance evaluation of an optimal cache replacement policy for wireless data 125 | P a g e