An introduction to Network-On-chips Sayed Amirhossein Mirhosseini - - PowerPoint PPT Presentation

an introduction to network on chips
SMART_READER_LITE
LIVE PREVIEW

An introduction to Network-On-chips Sayed Amirhossein Mirhosseini - - PowerPoint PPT Presentation

An introduction to Network-On-chips Sayed Amirhossein Mirhosseini Fully Crossbar Interconnection Commercial Example Niagara multiprocessor SoC from SUN BUS Buses are the simplest and most widely used SoC interconnection networks


slide-1
SLIDE 1

Sayed Amirhossein Mirhosseini

An introduction to Network-On-chips

slide-2
SLIDE 2

Fully Crossbar Interconnection

 Commercial Example

 Niagara multiprocessor SoC from SUN

slide-3
SLIDE 3

BUS

 Buses are the simplest and most widely used

SoC interconnection networks

 Definition: A collection of signals (wires) to

which one or more IP components (which need to communicate data with each other) are connected

 Only one IP component can transfer data on

the shared bus at any given time

slide-4
SLIDE 4

Shared Bus

 Commercial Example

 DaVinci family of digital video processing SoCs from

Texas Instruments

slide-5
SLIDE 5

Hierarchical Shared Bus

 Improves system throughput  Multiple ongoing transfers on different buses  Commercial example:

 Customizable multiprocessor ARM PrimeXsys SoCs

slide-6
SLIDE 6

Towards NOC

 Buses are not scalable

slide-7
SLIDE 7

Network Topologies

Torus Hypercube Binary Tree

slide-8
SLIDE 8

Topology Selection Factors

 Diameter:

 Maximum shortest path distance between two nodes in

the network. Networks with small diameters are preferable.

 Number of Links:

 A topology with large number of links can support high

bandwidth.

 Average Distance:

 Average among the distances between all pairs of nodes

  • f a graph. A topology having lesser average distance is

preferable.

slide-9
SLIDE 9

Topology Selection Factors (cont.)

 Node Degree:

 Numbers of channels connecting the node to its

  • neighbors. The lower this number, the easier to build

the network.

 Bisection Width:

 Minimum number of wires removed in order to bisect a

  • network. A larger bisection width enables faster

information exchange, and preferable.

slide-10
SLIDE 10

NOC Elements

 NoCs use packets to route data from the

source to the destination PE via a network fabric that consists of :

 switches (routers)  interconnection links (wires)

slide-11
SLIDE 11

Router Architecture

slide-12
SLIDE 12

Switching Strategies

 Determine how data flows through routers in

the network

 Flit :

 an elementary packet on which link flow control

  • perations are performed

 Phit :

 physical unit of data that is transferred on a link in a

single cycle

slide-13
SLIDE 13

Circuit Switching

 Physical path between the source and the

destination is reserved prior to the transmission of data

 Message’s header flit traverses the network

from the source to the destination, reserving links along the way

 When the probe reaches the destination, a

complete path has been set up and an acknowledgment is transmitted back to the source.

 Circuit switching is generally advantageous

when messages are infrequent and long

slide-14
SLIDE 14

Packet switching

 packets are transmitted from source and make

their way independently to receiver

 zero start up time, followed by a variable delay

due to contention in routers along packet path

slide-15
SLIDE 15

virtual cut-through switching

 The router can start forwarding the header

and following data bytes as soon as routing decisions have been made and the output buffer is free.

slide-16
SLIDE 16

Wormhole Switching

 flit from a packet is forwarded to receiving

router if space exists for that flit

 parts of the packet can be distributed among two or

more routers

 more susceptible to deadlocks due to usage

dependencies between links

slide-17
SLIDE 17

Deadlock

 One or more packets in a network become

blocked, and remain blocked for an indefinite amount of time waiting for an event that cannot occur.

 Example :

slide-18
SLIDE 18

Deadlock (cont.)

 Deadlock Avoidance : Techniques that

request resources in such a way that no deadlock can arise.

 Deadlock Recovery : These techniques require

a mechanism to detect and resolve potential deadlock situations.

slide-19
SLIDE 19

A Deadlock Free Router

slide-20
SLIDE 20

Livelock

 Livelocks are similar to deadlocks, except that

states of the resources involved constantly change with regard to one another, without making any progress

 occurs especially when dynamic (adaptive)

routing is used

 e.g. can occur in a deflective routing if a

packet is bounced around over and over again between routers and never reaches its destination

slide-21
SLIDE 21

Routing Algorithm

 Routing algorithms establish the path followed

by each message or packet .

 Many properties of the interconnection

network are a direct consequence of the routing algorithm used :

 Connectivity :

 Ability to route packets from any source node to any

destination node.  Adaptivity :

 Ability to route packets through alternative paths in

the presence of contention or faulty components.  Deadlock and livelock freedom :

 Ability to guarantee that packets will not block or

wander across the network forever.

slide-22
SLIDE 22

Routing Algorithms Taxonomy

slide-23
SLIDE 23

Deterministic Routing

 fixed paths are used to transfer data between

a particular source and destination

 Does not take into account current state of

the network

 Advantages :

 easy to implement  in-order packet delivery if single path is used  Manageable to be deadlock and livelock free

slide-24
SLIDE 24

Adaptive Routing

 routing decisions are made according to the

current state of the network

 Considering factors such as availability and

load on links

 Path between source and destination may

change over time

 More resources needed to monitor state of

the network and dynamically change routing paths

 Able to better distribute traffic in a

network

slide-25
SLIDE 25

Example : Odd-Even Routing (cont.)

 The odd-even turn model is governed by the

following two rules:

1.

Any packet is not allowed to take an EN turn at any nodes located in an even column, and it is not allowed to take an NW turn at any nodes located in an odd column.

  • 2. Any packet is not allowed to take an ES turn at

any nodes located in an even column, and it is not allowed to take an SW turn at any nodes located in an odd column.

slide-26
SLIDE 26

Example : Odd-Even Routing (cont.)

 Theorem : Any routing algorithms that follow

the rules of the odd-even turn model are deadlock free as long as 180-degree turns are prohibited.

 Proof :

 By Contradiction :

NOT Possible !

slide-27
SLIDE 27

Example : Odd-Even Routing (cont.)

slide-28
SLIDE 28

Virtual channel

 Buffers at physical channels are commonly

  • perated as FIFO queues. Therefore, once a

message occupies a buffer for a channel, no other message can access the physical channel.

 Each virtual channel is realized by an

independently managed pair of message buffers.

slide-29
SLIDE 29

A Virtual Channeled Router

slide-30
SLIDE 30

Many ways to gain bonus marks 

 Example : Pipelined Router

slide-31
SLIDE 31

References

 Principles and Practices of Interconnection

Networks by William James Dally and Brian Pattrick Towles [2004]

 Interconnection Networks: An Engineering

Approach by Jose Duato, Sudhakar Yalamanchili and Lionel M. Ni [1997]

 Computer Architecture, Fifth Edition: A

Quantitative Approach by John L.Hennessy and David A. Patterson [2011]

 Chiu, Ge-Ming. "The odd-even turn model for

adaptive routing." Parallel and Distributed Systems, IEEE Transactions on 11.7 (2000): 729-738.