INTERCONNECTION NETWORKS

MOTIVATION : CONCURRENT PROCESSING

(e.g. one billion flop instructions
per second )

Program (User / Application)

Task Partition ‘ DIVIDE – AND – CONVERT ‘

♦ INDIVISIBLE
♦ MODULARLY DIVISIBLE
♦ ARBITRARILY DIVISIBLE

.....
P1 Pn
Process 1. . . . . . . . . . . Process n

I. N.

m1 m2 . . . . . . . mk

Architecture choice : a) Operation

b) Control Strategy
c) Switching Strategy
d) Network Topology

Operation : Asynchronous & Synchronous

Control Strategy : Setting Control on Switching

Elements
(Centralized or De-Centralized)

Switching Strategy : Ckt. & Pkt. Switching

NETWORK TOPOLOGY : a. Static
Based on Re-configuration
b. Dynamic

S = Space for Interconnection Networks evolves due to

{Operation Mode} X {Control Strategy}X {Switching Strategy} X {Network Strategy}

♦ What is important in terms of performance is in the choice of an apt topology

for different applications.

STATIC

Linear Array, Star, Bus, Ring, Tree, Mesh, Systolic Arrays, Complete Graphs, n-
cube

Application System Topology

Performance Criteria

Dimensional Aspect : one, two and three....... are possible

DYNAMIC a) Single Stage b) Multi Stage c) Cross Bar

a) b) c)
1
2

3 Multiple
4 Stages

Single Stage : e.g. Shuffle Exchange Network –

Data may have to recirculate many times before it can reach
its destination (recirculating networks
Multi Stage : Capable of connecting an arbitrary input terminal to an arbitrary
Output terminal.

This can be a) One – Sided b) Two – Sided

Input & Output Input & Output

on the same side on two sides

Blocking Re-arrangeable Non- Blocking

Blocking : Simultaneous connections of more than one terminal pair may

result in conflicts in the use of network resources ( Common
links, etc..) eg. Baseline, Omega, Binary n-cube

Rearrangeable : Referred to as “rearrangeably non-blocking “

It can perform all possible connections between inputs and
Outputs by rearranging the existing connections
eg. Benes network

Non-Blocking : This can handle all possible connections without blocking

eg. Clos Networks

OTHER ISSUES : Communication Protocols, Routing Techniques, Data Distbn.

Strategy(Scheduling Policies), Reliability and Fault Tolerance.
 SHUFFLE EXCHANGE NETWORKS

Principle : Based on perfect shuffling of playing cards .

The interconnection pattern is as follows :

0 0 0 1 2 3 4 5 6 7
1 1

2 2
3 3 0 4 1 5 2 6 3 7

4 4
5 5

6 6 0 1 0
7 7

Perfect Shuffle Pattern

1 0 0
i.e.2 4

Shuffle (xn-1, xn-2,........ x1, x0) = xn-2.........x1,x0,xn-1

Exchange (xn-1, xn-2,........ x1, x0) = xn-1.........x1,x0

Meant for re-circulation until desired connection is achieved

 BASE LINE NETWORK

* - Self-routing ability

000 000
001 001

010 010
011 011

100 100
101 101

110 110
111 111

stage 1(MSB) stage 3(LSB)

Self- routing based on address tags

For each stage we have a switching control defined by a bit position

Stage 1 (MSB) to Stage N(LSB)

Switchable for Pkt. Switch Networks

Self-routing optical networks are quite popular

(Use of CDMA Techniques for Self-routing) -V

Destination : 1 1’
2 2’

0 1 0 +V
0 – Up ; 1 - Down
 CLOS NETWORK

r1 X r2 m X n2
1 1 1 1 1 1
2 2 2 2 2

1 1

n1 r1 r 2 m n2

n1+1 1 1 1 1 n2+1
2 2 2 2

2 2

2n1 r 1 r2 m 2n2

1 1 1 1
2 2 2 2

m r2

r1n1 m r1 r2 m r2n2

Any one input has a parth to any output


Non- Blocking or not depends on the # of elements in the middle stage

Lemma : Non-Blocking, if m ≥ n2 + n1 – 1 where m is the # of middle

Stage switching elements