SlideShare a Scribd company logo
Matthew Agostinelli Interconnect Technology 5/2/2014
1
Introduction
Computer processors in this world continue to run
faster and more efficient as the technology for them
evolve. There is a common goal for processors to be
able to run a plethora of operations but use less
power. With changing technology, the interconnect
network of a processor becomes more of a problem
as transistor size becomes smaller. The power
dissipation alone with current Interconnect
technology accounts for more than 20 percent of
power used in a processor. In this article there will be
detailed description of the problems that need to be
overcome to make processors more efficient and
powerful. In this article we will explore the different
ways the interconnect network of processors have
evolved over time and the different techniques used
to design the most efficient processors of today. A
general rule of thumb for the total cost of a
supercomputer is that two thirds of the total cost is
due to the processor and memory modules and one
third of the cost is the network itself. [1] So the
development of new technology in processor design
is a huge topic of discussion as companies like
AMD®, and Intel® are in the race to design the
fastest most efficient processor.
Cost of the interconnection network
In today’s world one of the main goals for computer
design is to design a cheaper computer. The most
crucial part of the computer would be the processor.
The processor has a very complex interconnect
network. It is estimated that as much as one third of
the network cost is related to the total number of
wires; this cost includes the expense of driving
messages at very high rates through the wires. [1] So
in turn developing a better interconnect network
make a cheaper processor which would be a huge
step in the right direction.
Materials used
For the longest time aluminum metal was used for the
wiring of a computer processor. It is a lightweight
metal and doesn’t react with the silicon in the wafer
when bonded to it. The problem with the aluminum
wiring is the fact that it was very expensive to use
and hard to design complicated interconnect
infrastructures due to wire diameter. IBM in 1997
introduced a new technology which was thought to
be impossible. They introduced the first copper chip.
Copper wires conduct electricity with about 40
percent less resistance than aluminum wires, which
results in an additional 15 percent burst in
microprocessor speed. Copper wires are also
significantly more durable and 100 times more
reliable over time, and can be shrunk to smaller sizes
than aluminum. [4] Not only was the copper more
efficient, it made it possible to lay more complicated
wire networks which it turn made the processors
more powerful. One drawback to the copper wiring
was how it reacted with the silicon in the wafer. It
alters the electrical properties of the silicon when in
contact with the wafer. A solution to this problem
was to create an insulating layer in between the
silicon and the copper wiring. Over time there are
numerous advancements in this technology as
processors continue to evolve with engineering
standards.
Switching Analogies
Ways to increase processor speeds would be to create
better switching methods between the header node,
memory, and the destination node. We want low
buffering space to be used, and latency of the
message being sent to be as small as possible.
Latency is defined in clock cycles, and is the time
from header injection into the network then finally
into the destination node. Since there are many
messages being sent from different nodes the average
latency must be defined.
𝐿𝑖 = 𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 𝐿𝑎𝑡𝑒𝑛𝑐𝑦 𝑜𝑓 𝑎 𝑚𝑒𝑠𝑠𝑎𝑔𝑒
𝑃 = 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑒𝑠𝑠𝑎𝑔𝑒𝑠
𝐿 𝑎𝑣𝑔 =
𝐿𝑖
𝑃
Buffering Space is used for saving or loading
statements that need to be sent.
Circuit Switching
With circuit switching there is a defined path which
is reserved for the transmission of data. This path is
reserved until all data has been transmitted.
Disadvantages to this approach, is that valuable
resources are held up while data is being transferred.
This in turn causes unnecessary delays in the path.
Packet Switching
In this method data is divided into fixed length
blocks, there is no defined path of the packet being
sent. When the source has a packet to be sent is
transmits the data. This form of switching requires a
large buffer requirement which isn’t helpful when
designing a fast processor.
Matthew Agostinelli Interconnect Technology 5/2/2014
2
Wormhole switching
This form of switching is generally used for
interconnect design. For starters the switches can be
small and compact making it easier to design more
complicated networks. The way this technique works
is it sends out a header flit which contains the routing
information. The flit follows this path in a pipeline
fashion. The subsequent flits then follow the path
defined until all data is received. One drawback to
this method would be that messages have to cross
their channel entirely before the channel can be used
for another message. One way to fix this is to make
virtual channels in the input and output ports which
take flits belonging to a particular packet and allow
flits to use the virtual channel buffers. This increases
the utilization of the physical channel. Our goal with
the switching architecture is to have high throughput.
The operation of a switch has three processes. First it
has an input arbitration, then routing is defined, and
finally an output arbitration. As mentioned above to
have high throughput we need to have a virtual
channel switch where each port has multiple parallel
buffers.
Energy dissipation
With a complex interconnect network energy
dissipation becomes a big issue as complexity of the
network increases. The total energy dissipated is a
combination of energy from switches and energy
from the wiring network.
𝐸 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑜𝑟 = Σ𝐴𝐹𝑗 𝑪𝒋 𝑉2
𝑓
𝐸 = 𝐸𝑠𝑤𝑖𝑡𝑐ℎ + 𝐸𝑖𝑛𝑡𝑒𝑟𝑐𝑜𝑛𝑛𝑒𝑐𝑡
The energy from the switches and interconnect
depend on the capacitance from the signal activity
𝐸 = ∝ 𝑠𝑤𝑖𝑡𝑐ℎ∗ 𝑪𝑉2
To reduce the energy dissipated simply just reduce
the wire length of the interconnect infrastructure.
Architectures that possess longer interswitch wires
will generally create more routing challenges [4]. The
less wire used would decrease the energy dissipation.
Switching algorithms could reduce Energy
dissipation and can be referred to with routing
algorithms later in the article. It is said that about
50% percent of the power dissipated in a processor is
due to the interconnect network. Yet only 10% of
interconnect power dissipation is due to the routing
network. The switches at each junction have a
capacitance and as stated above energy dissipation is
directly correlated to the capacitance of the switches
so one way to reduce this energy dissipation is to
develop better routing algorithms which reduce
overall capacitance for the network.
Routing algorithms
One way to make computer processors more efficient
is to develop better routing algorithms to send data.
In a processor most of the power dissipation is due to
the interconnect network. We want to reduce the
power consumption and one way to do that is to
reduce wire capacitance - simple and efficient routing
algorithms, small diameter, high connectivity, and
small degree. Also, one would wish the
interconnection network to be as efficient as possible.
[1]
Reduce Wire Length
One way to reduce wire capacitance would be to
reduce overall length of the wire. Since the energy
dissipated is directly related to capacitance in the
wire making it shorter would reduce power
consumption overall.
Interconnect Spacing
In the majority of processors most networks are
routed at minimal spacing to reduce materials cost.
This isn’t helpful because as wire spacing decreases
the power dissipation and wire capacitance increase.
The graph below shows how spacing and capacitance
are related.
Figure 1: Taken from [7]
Conclusion
In our world today we want to have the most
powerful processors on the market. There is currently
Matthew Agostinelli Interconnect Technology 5/2/2014
3
a huge push to develop software that doesn’t require
so much processing power. This is good because
developing cheaper processors that run simpler
software make overall device prices a lot cheaper.
The goal used to be high bandwidth and high
throughput but with the new software developments
the focus on infrastructure would be less power
consumption. With mobile processors in tablets and
cellphones the main goal would be exactly that.
Energy dissipation is mainly due to the switches in
the interconnect network. Routing algorithms are
being developed to find solutions to this problem.
Reducing wire length and using the right spacing
could dramatically reduce power consumption. The
interconnect infrastructure of a computer processor is
very important and extensively researched. More
development in this field could lead to a better future
in mobile computing.
Bibliography
[1] Schibell, Stephen T., and Richard M. Stafford.
"Processor Interconnection Networks from Cayley
Graphs." Discrete Applied Mathmatics: 35-55. Print.
[2] Pande, P. P., et al. "Performance evaluation and
design trade-offs for network-on-chip interconnect
architectures." IEEE TRANSACTIONS ON
COMPUTERS 54.8 (2005): 1025-40. Print.
[3] Wentzlaff, David, et al. "On-chip Interconnection
Architechture of the Tile Processor." IEEE Computer
Society (2007): n. pag. Berkeley.edu. Web. 1 May
2014. <http://www.cs.berkeley.edu/
~kubitron/cs258/handouts/papers/wentzlaff_tile64_n
oc_ieeemicro2007.pdf>.
[4] Issac, Randy. "Copper Interconnects The
Evolution of Microprocessors." Intel. Intel, n.d. Web.
1 May 2014. <http://www-
03.ibm.com/ibm/history/ibm100/us/en/icons/copperc
hip/>.
[5] "An Introduction to the Intel ® QuickPath
Interconnect." Intel. Intel, Jan. 2009. Web. 1 May
2014. <http://www.intel.com/content/dam/doc/white-
paper/ quick-path-interconnect-introduction-
paper.pdf>.
[6] Palesi, M.; Holsmark, R.; Kumar, S.; Catania, V.
"Application Specific Routing Algorithms for
Networks on Chip", Parallel and Distributed Systems,
IEEE Transactions on, On page(s): 316 - 330
Volume: 20, Issue: 3, March 2009
[7] Mangen, Nir, et al. "Interconnect-Power
Dissipation in a Microprocessor." SLIP (2004): 7-13.
Print.
Matthew Agostinelli Interconnect Technology 5/2/2014
4

More Related Content

What's hot

Aq4103266271
Aq4103266271Aq4103266271
Aq4103266271
IJERA Editor
 
TYPES OF COMPUTER NETWORK CABLES
TYPES OF COMPUTER NETWORK CABLES TYPES OF COMPUTER NETWORK CABLES
TYPES OF COMPUTER NETWORK CABLES
Sayma Sultana
 
Energy Minimization in Wireless Sensor Networks Using Multi Hop Transmission
Energy Minimization in Wireless Sensor Networks Using Multi  Hop TransmissionEnergy Minimization in Wireless Sensor Networks Using Multi  Hop Transmission
Energy Minimization in Wireless Sensor Networks Using Multi Hop Transmission
IOSR Journals
 
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
chokrio
 
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
ijgca
 
Green Networks
Green NetworksGreen Networks
Green Networks
Neenu Ks
 
Joint beamforming, power and channel allocation in multi user and multi-chann...
Joint beamforming, power and channel allocation in multi user and multi-chann...Joint beamforming, power and channel allocation in multi user and multi-chann...
Joint beamforming, power and channel allocation in multi user and multi-chann...
Pvrtechnologies Nellore
 
Topology
TopologyTopology
Topology
MdAnikKhan
 
Network cable types and specifications
Network cable types and specificationsNetwork cable types and specifications
Network cable types and specifications
Maksudujjaman
 
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
ijasuc
 
Block diagonalization precoding and power allocation for clustering small-cel...
Block diagonalization precoding and power allocation for clustering small-cel...Block diagonalization precoding and power allocation for clustering small-cel...
Block diagonalization precoding and power allocation for clustering small-cel...
journalBEEI
 
Ftp and database statistics in wireless network environment for web client 2
Ftp and database statistics in wireless network environment for web client 2Ftp and database statistics in wireless network environment for web client 2
Ftp and database statistics in wireless network environment for web client 2
IAEME Publication
 
Reducing Energy Consumption in LTE with Cell DTX
Reducing Energy Consumption in LTE with Cell DTX Reducing Energy Consumption in LTE with Cell DTX
Reducing Energy Consumption in LTE with Cell DTX
Prashant Panigrahi
 
An Energy Efficient Protocol To Increase Network Life In WSN
An Energy Efficient Protocol To Increase Network Life In WSNAn Energy Efficient Protocol To Increase Network Life In WSN
An Energy Efficient Protocol To Increase Network Life In WSN
IOSR Journals
 
Energy saving in cooperative transmission using opportunistic protocol in MANET
Energy saving in cooperative transmission using opportunistic protocol in MANETEnergy saving in cooperative transmission using opportunistic protocol in MANET
Energy saving in cooperative transmission using opportunistic protocol in MANET
IOSR Journals
 
Note 1
Note 1Note 1
Computer Networks 2
Computer Networks 2Computer Networks 2
Computer Networks 2
Mr Smith
 
Cables and connectors of computer networks
Cables and connectors of computer networksCables and connectors of computer networks
Cables and connectors of computer networks
kona paul
 

What's hot (18)

Aq4103266271
Aq4103266271Aq4103266271
Aq4103266271
 
TYPES OF COMPUTER NETWORK CABLES
TYPES OF COMPUTER NETWORK CABLES TYPES OF COMPUTER NETWORK CABLES
TYPES OF COMPUTER NETWORK CABLES
 
Energy Minimization in Wireless Sensor Networks Using Multi Hop Transmission
Energy Minimization in Wireless Sensor Networks Using Multi  Hop TransmissionEnergy Minimization in Wireless Sensor Networks Using Multi  Hop Transmission
Energy Minimization in Wireless Sensor Networks Using Multi Hop Transmission
 
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
NEW APPROACH TO IMPROVING LIFETIME IN HETEROGENEOUS WIRELESS SENSOR NETWORKS ...
 
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
ON THE CELL BREATHING TECHNIQUE TO REDUCE CONGESTION APPLYING BANDWIDTH LIMIT...
 
Green Networks
Green NetworksGreen Networks
Green Networks
 
Joint beamforming, power and channel allocation in multi user and multi-chann...
Joint beamforming, power and channel allocation in multi user and multi-chann...Joint beamforming, power and channel allocation in multi user and multi-chann...
Joint beamforming, power and channel allocation in multi user and multi-chann...
 
Topology
TopologyTopology
Topology
 
Network cable types and specifications
Network cable types and specificationsNetwork cable types and specifications
Network cable types and specifications
 
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
EBCD: A ROUTING ALGORITHM BASED ON BEE COLONY FOR ENERGY CONSUMPTION REDUCTIO...
 
Block diagonalization precoding and power allocation for clustering small-cel...
Block diagonalization precoding and power allocation for clustering small-cel...Block diagonalization precoding and power allocation for clustering small-cel...
Block diagonalization precoding and power allocation for clustering small-cel...
 
Ftp and database statistics in wireless network environment for web client 2
Ftp and database statistics in wireless network environment for web client 2Ftp and database statistics in wireless network environment for web client 2
Ftp and database statistics in wireless network environment for web client 2
 
Reducing Energy Consumption in LTE with Cell DTX
Reducing Energy Consumption in LTE with Cell DTX Reducing Energy Consumption in LTE with Cell DTX
Reducing Energy Consumption in LTE with Cell DTX
 
An Energy Efficient Protocol To Increase Network Life In WSN
An Energy Efficient Protocol To Increase Network Life In WSNAn Energy Efficient Protocol To Increase Network Life In WSN
An Energy Efficient Protocol To Increase Network Life In WSN
 
Energy saving in cooperative transmission using opportunistic protocol in MANET
Energy saving in cooperative transmission using opportunistic protocol in MANETEnergy saving in cooperative transmission using opportunistic protocol in MANET
Energy saving in cooperative transmission using opportunistic protocol in MANET
 
Note 1
Note 1Note 1
Note 1
 
Computer Networks 2
Computer Networks 2Computer Networks 2
Computer Networks 2
 
Cables and connectors of computer networks
Cables and connectors of computer networksCables and connectors of computer networks
Cables and connectors of computer networks
 

Similar to Interconnect technology Final Revision

C017631521
C017631521C017631521
C017631521
IOSR Journals
 
Review on Green Networking Solutions
Review on Green Networking SolutionsReview on Green Networking Solutions
Review on Green Networking Solutions
iosrjce
 
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
IJCNCJournal
 
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
Editor IJCATR
 
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
ijceronline
 
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
IJCNCJournal
 
Lan Switching[1]
Lan Switching[1]Lan Switching[1]
Lan Switching[1]
sarvodaya2001
 
Pid967241
Pid967241Pid967241
Pid967241
Shahab Shahid
 
Distributed compressive sampling for lifetime
Distributed compressive sampling for lifetimeDistributed compressive sampling for lifetime
Distributed compressive sampling for lifetime
topssy
 
An Opportunistic Routing Protocol
An Opportunistic Routing ProtocolAn Opportunistic Routing Protocol
An Opportunistic Routing Protocol
dbpublications
 
Study of Energy Saving in Carrier-Ethernet Network
Study of Energy Saving in Carrier-Ethernet NetworkStudy of Energy Saving in Carrier-Ethernet Network
Study of Energy Saving in Carrier-Ethernet Network
cleberaraujo
 
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
eSAT Journals
 
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
eSAT Publishing House
 
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder CircuitDesign and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
Associate Professor in VSB Coimbatore
 
Chapter 8 the role of networking in manufacturing
Chapter 8   the role of networking in manufacturingChapter 8   the role of networking in manufacturing
Chapter 8 the role of networking in manufacturing
N. A. Sutisna
 
Congestion aware routing algorithm network on chip
Congestion aware routing algorithm network on chipCongestion aware routing algorithm network on chip
Congestion aware routing algorithm network on chip
NiteshKumar198644
 
Crosslayertermpaper
CrosslayertermpaperCrosslayertermpaper
Crosslayertermpaper
B.T.L.I.T
 
“Optimizing the data transmission between multiple nodes during link failure ...
“Optimizing the data transmission between multiple nodes during link failure ...“Optimizing the data transmission between multiple nodes during link failure ...
“Optimizing the data transmission between multiple nodes during link failure ...
eSAT Publishing House
 
Low power network on chip architectures: A survey
Low power network on chip architectures: A surveyLow power network on chip architectures: A survey
Low power network on chip architectures: A survey
CSITiaesprime
 
ENERGY EFFICIENT ROUTING ALGORITHM
ENERGY EFFICIENT ROUTING ALGORITHMENERGY EFFICIENT ROUTING ALGORITHM

Similar to Interconnect technology Final Revision (20)

C017631521
C017631521C017631521
C017631521
 
Review on Green Networking Solutions
Review on Green Networking SolutionsReview on Green Networking Solutions
Review on Green Networking Solutions
 
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
ENERGY SAVINGS IN APPLICATIONS FOR WIRELESS SENSOR NETWORKS TIME CRITICAL REQ...
 
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
A New Method for Reducing Energy Consumption in Wireless Sensor Networks usin...
 
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
An Approach for Enhanced Performance of Packet Transmission over Packet Switc...
 
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
AN EFFICIENT BANDWIDTH OPTIMIZATION AND MINIMIZING ENERGY CONSUMPTION UTILIZI...
 
Lan Switching[1]
Lan Switching[1]Lan Switching[1]
Lan Switching[1]
 
Pid967241
Pid967241Pid967241
Pid967241
 
Distributed compressive sampling for lifetime
Distributed compressive sampling for lifetimeDistributed compressive sampling for lifetime
Distributed compressive sampling for lifetime
 
An Opportunistic Routing Protocol
An Opportunistic Routing ProtocolAn Opportunistic Routing Protocol
An Opportunistic Routing Protocol
 
Study of Energy Saving in Carrier-Ethernet Network
Study of Energy Saving in Carrier-Ethernet NetworkStudy of Energy Saving in Carrier-Ethernet Network
Study of Energy Saving in Carrier-Ethernet Network
 
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
 
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...Performance improvement of bottleneck link in red vegas over heterogeneous ne...
Performance improvement of bottleneck link in red vegas over heterogeneous ne...
 
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder CircuitDesign and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
Design and Analysis of Low Power High Speed Hybrid logic 8-T Full Adder Circuit
 
Chapter 8 the role of networking in manufacturing
Chapter 8   the role of networking in manufacturingChapter 8   the role of networking in manufacturing
Chapter 8 the role of networking in manufacturing
 
Congestion aware routing algorithm network on chip
Congestion aware routing algorithm network on chipCongestion aware routing algorithm network on chip
Congestion aware routing algorithm network on chip
 
Crosslayertermpaper
CrosslayertermpaperCrosslayertermpaper
Crosslayertermpaper
 
“Optimizing the data transmission between multiple nodes during link failure ...
“Optimizing the data transmission between multiple nodes during link failure ...“Optimizing the data transmission between multiple nodes during link failure ...
“Optimizing the data transmission between multiple nodes during link failure ...
 
Low power network on chip architectures: A survey
Low power network on chip architectures: A surveyLow power network on chip architectures: A survey
Low power network on chip architectures: A survey
 
ENERGY EFFICIENT ROUTING ALGORITHM
ENERGY EFFICIENT ROUTING ALGORITHMENERGY EFFICIENT ROUTING ALGORITHM
ENERGY EFFICIENT ROUTING ALGORITHM
 

Interconnect technology Final Revision

  • 1. Matthew Agostinelli Interconnect Technology 5/2/2014 1 Introduction Computer processors in this world continue to run faster and more efficient as the technology for them evolve. There is a common goal for processors to be able to run a plethora of operations but use less power. With changing technology, the interconnect network of a processor becomes more of a problem as transistor size becomes smaller. The power dissipation alone with current Interconnect technology accounts for more than 20 percent of power used in a processor. In this article there will be detailed description of the problems that need to be overcome to make processors more efficient and powerful. In this article we will explore the different ways the interconnect network of processors have evolved over time and the different techniques used to design the most efficient processors of today. A general rule of thumb for the total cost of a supercomputer is that two thirds of the total cost is due to the processor and memory modules and one third of the cost is the network itself. [1] So the development of new technology in processor design is a huge topic of discussion as companies like AMD®, and Intel® are in the race to design the fastest most efficient processor. Cost of the interconnection network In today’s world one of the main goals for computer design is to design a cheaper computer. The most crucial part of the computer would be the processor. The processor has a very complex interconnect network. It is estimated that as much as one third of the network cost is related to the total number of wires; this cost includes the expense of driving messages at very high rates through the wires. [1] So in turn developing a better interconnect network make a cheaper processor which would be a huge step in the right direction. Materials used For the longest time aluminum metal was used for the wiring of a computer processor. It is a lightweight metal and doesn’t react with the silicon in the wafer when bonded to it. The problem with the aluminum wiring is the fact that it was very expensive to use and hard to design complicated interconnect infrastructures due to wire diameter. IBM in 1997 introduced a new technology which was thought to be impossible. They introduced the first copper chip. Copper wires conduct electricity with about 40 percent less resistance than aluminum wires, which results in an additional 15 percent burst in microprocessor speed. Copper wires are also significantly more durable and 100 times more reliable over time, and can be shrunk to smaller sizes than aluminum. [4] Not only was the copper more efficient, it made it possible to lay more complicated wire networks which it turn made the processors more powerful. One drawback to the copper wiring was how it reacted with the silicon in the wafer. It alters the electrical properties of the silicon when in contact with the wafer. A solution to this problem was to create an insulating layer in between the silicon and the copper wiring. Over time there are numerous advancements in this technology as processors continue to evolve with engineering standards. Switching Analogies Ways to increase processor speeds would be to create better switching methods between the header node, memory, and the destination node. We want low buffering space to be used, and latency of the message being sent to be as small as possible. Latency is defined in clock cycles, and is the time from header injection into the network then finally into the destination node. Since there are many messages being sent from different nodes the average latency must be defined. 𝐿𝑖 = 𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 𝐿𝑎𝑡𝑒𝑛𝑐𝑦 𝑜𝑓 𝑎 𝑚𝑒𝑠𝑠𝑎𝑔𝑒 𝑃 = 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑚𝑒𝑠𝑠𝑎𝑔𝑒𝑠 𝐿 𝑎𝑣𝑔 = 𝐿𝑖 𝑃 Buffering Space is used for saving or loading statements that need to be sent. Circuit Switching With circuit switching there is a defined path which is reserved for the transmission of data. This path is reserved until all data has been transmitted. Disadvantages to this approach, is that valuable resources are held up while data is being transferred. This in turn causes unnecessary delays in the path. Packet Switching In this method data is divided into fixed length blocks, there is no defined path of the packet being sent. When the source has a packet to be sent is transmits the data. This form of switching requires a large buffer requirement which isn’t helpful when designing a fast processor.
  • 2. Matthew Agostinelli Interconnect Technology 5/2/2014 2 Wormhole switching This form of switching is generally used for interconnect design. For starters the switches can be small and compact making it easier to design more complicated networks. The way this technique works is it sends out a header flit which contains the routing information. The flit follows this path in a pipeline fashion. The subsequent flits then follow the path defined until all data is received. One drawback to this method would be that messages have to cross their channel entirely before the channel can be used for another message. One way to fix this is to make virtual channels in the input and output ports which take flits belonging to a particular packet and allow flits to use the virtual channel buffers. This increases the utilization of the physical channel. Our goal with the switching architecture is to have high throughput. The operation of a switch has three processes. First it has an input arbitration, then routing is defined, and finally an output arbitration. As mentioned above to have high throughput we need to have a virtual channel switch where each port has multiple parallel buffers. Energy dissipation With a complex interconnect network energy dissipation becomes a big issue as complexity of the network increases. The total energy dissipated is a combination of energy from switches and energy from the wiring network. 𝐸 𝑝𝑟𝑜𝑐𝑒𝑠𝑠𝑜𝑟 = Σ𝐴𝐹𝑗 𝑪𝒋 𝑉2 𝑓 𝐸 = 𝐸𝑠𝑤𝑖𝑡𝑐ℎ + 𝐸𝑖𝑛𝑡𝑒𝑟𝑐𝑜𝑛𝑛𝑒𝑐𝑡 The energy from the switches and interconnect depend on the capacitance from the signal activity 𝐸 = ∝ 𝑠𝑤𝑖𝑡𝑐ℎ∗ 𝑪𝑉2 To reduce the energy dissipated simply just reduce the wire length of the interconnect infrastructure. Architectures that possess longer interswitch wires will generally create more routing challenges [4]. The less wire used would decrease the energy dissipation. Switching algorithms could reduce Energy dissipation and can be referred to with routing algorithms later in the article. It is said that about 50% percent of the power dissipated in a processor is due to the interconnect network. Yet only 10% of interconnect power dissipation is due to the routing network. The switches at each junction have a capacitance and as stated above energy dissipation is directly correlated to the capacitance of the switches so one way to reduce this energy dissipation is to develop better routing algorithms which reduce overall capacitance for the network. Routing algorithms One way to make computer processors more efficient is to develop better routing algorithms to send data. In a processor most of the power dissipation is due to the interconnect network. We want to reduce the power consumption and one way to do that is to reduce wire capacitance - simple and efficient routing algorithms, small diameter, high connectivity, and small degree. Also, one would wish the interconnection network to be as efficient as possible. [1] Reduce Wire Length One way to reduce wire capacitance would be to reduce overall length of the wire. Since the energy dissipated is directly related to capacitance in the wire making it shorter would reduce power consumption overall. Interconnect Spacing In the majority of processors most networks are routed at minimal spacing to reduce materials cost. This isn’t helpful because as wire spacing decreases the power dissipation and wire capacitance increase. The graph below shows how spacing and capacitance are related. Figure 1: Taken from [7] Conclusion In our world today we want to have the most powerful processors on the market. There is currently
  • 3. Matthew Agostinelli Interconnect Technology 5/2/2014 3 a huge push to develop software that doesn’t require so much processing power. This is good because developing cheaper processors that run simpler software make overall device prices a lot cheaper. The goal used to be high bandwidth and high throughput but with the new software developments the focus on infrastructure would be less power consumption. With mobile processors in tablets and cellphones the main goal would be exactly that. Energy dissipation is mainly due to the switches in the interconnect network. Routing algorithms are being developed to find solutions to this problem. Reducing wire length and using the right spacing could dramatically reduce power consumption. The interconnect infrastructure of a computer processor is very important and extensively researched. More development in this field could lead to a better future in mobile computing. Bibliography [1] Schibell, Stephen T., and Richard M. Stafford. "Processor Interconnection Networks from Cayley Graphs." Discrete Applied Mathmatics: 35-55. Print. [2] Pande, P. P., et al. "Performance evaluation and design trade-offs for network-on-chip interconnect architectures." IEEE TRANSACTIONS ON COMPUTERS 54.8 (2005): 1025-40. Print. [3] Wentzlaff, David, et al. "On-chip Interconnection Architechture of the Tile Processor." IEEE Computer Society (2007): n. pag. Berkeley.edu. Web. 1 May 2014. <http://www.cs.berkeley.edu/ ~kubitron/cs258/handouts/papers/wentzlaff_tile64_n oc_ieeemicro2007.pdf>. [4] Issac, Randy. "Copper Interconnects The Evolution of Microprocessors." Intel. Intel, n.d. Web. 1 May 2014. <http://www- 03.ibm.com/ibm/history/ibm100/us/en/icons/copperc hip/>. [5] "An Introduction to the Intel ® QuickPath Interconnect." Intel. Intel, Jan. 2009. Web. 1 May 2014. <http://www.intel.com/content/dam/doc/white- paper/ quick-path-interconnect-introduction- paper.pdf>. [6] Palesi, M.; Holsmark, R.; Kumar, S.; Catania, V. "Application Specific Routing Algorithms for Networks on Chip", Parallel and Distributed Systems, IEEE Transactions on, On page(s): 316 - 330 Volume: 20, Issue: 3, March 2009 [7] Mangen, Nir, et al. "Interconnect-Power Dissipation in a Microprocessor." SLIP (2004): 7-13. Print.
  • 4. Matthew Agostinelli Interconnect Technology 5/2/2014 4