A comparative study of Star and Arrangement

graph in Interconnection Networks

Greeshma P G

Guided by: Ajish K S

Reg No :TCR18CSCE05
Table of contents

1. Introduction

2. Star Graph

3. Arrangement Graph

4. Conclusion

5. References

1
Introduction
Parellel Computers

In Parellel Computers a problem is broken into a number of

subproblems. All of these subproblems are solved simultaneously,
each on a different processor. In doing so, processors may
communicate with one another to exchange partial results. Finally,
the results are combined to produce an answer to the original
problem.
A parallel computer solves a problem by utilizing several processors.

2
Contd..

Conceptually employing several processors to work together on a

given computation terminates in a totally new paradigm in computer
problem solving. It offers precious techniques for the design and
analysis of algorithms.

Parallel computers are classified into two major categories, namely,

shared-memory parallel machines and interconnection networks ,
depending on how the processors communicate with each other.

3
Shared-Memory parallel machines

• The class of shared-memory parallel computers is also known

as the Parallel Random Access Machine (PRAM).
• It consists of a number of identical processors P1 , P2 , ... , Pn and
a common memory which is shared by these n processors.

4
Contd..

Figure 1: shared memory parellel machine

5
Interconnection Networks

• In an interconnection network, there is no longer a shared

memory; instead, each processor has its own local memory and
connects with other processors via direct links between them.
• The links are two-way communication lines; in other words, two
processors connected by a link can exchange data
simultaneously.
• Mathematically, an undirected graph G = (V, E) can be used to
describe an interconnection network, where each processor Pi is
a vertex in V and if there is a link between two processors Pi and
Pj in the interconnection network, then an edge (Pi , Pj ) ∈ E
exists between the two responding vertices in the graph
• Two processors directly connected by a link are said to be
neighbours

6
Contd..

Figure 2: Interconnection Network parellel machine

7
Contd..

Terminolgies in Interconnection Networks

• Degree
The degree of a processor is the number of neighbours of this
processor. The degree of network topology is the maximum of
all processors’ degrees in the network
• Distance
The distance between two processors Pm and Pn is the number
of links on the shortest path from Pm to Pn
• Diameter
The diameter of network is the maximum distance among any
two arbitrary processors
• An efficient interconnection topology would usually require
small diameter and low degree
• Connectivity
The connectivity of a graph G is the smallest number of vertices
8
we can delete in order to disconnect G
Contd..

• A regular graph means that all nodes in this graph have the
same degree
• Isomorphism
An isomorphism from a graph G to a graph H is a bijection
f : V(G) → V(H) such that (u,v) E(G) if and only if (f(u),f(v)) ∈ E(H)
• G is isomorphic to H if there is an isomorphism from G to H
• An automorphism of a graph is an isomorphism from G to G.
• Node(Edge)-symmetric
A graph is node-symmetric (edge-symmetric) if for every pair of
nodes u , v ∈ V (edges e , f ∈ E(G)), there is an automorphism
that maps u to v (e to f )
• This means that in a symmetric graph, if you look at the whole
graph from any node (edge) of V (G) (E(G)) it will be exactly the
same graph

9
Contd..

• Fault tolerence
A graph G is called fault tolerant when the removal of f or less
nodes does not make the graph disconnected
• This indicates the maximum number of faulty or blocked nodes
that can exist in a graph, but the network is still able to continue.

There exist many interconnection networks, each performing better

than the other ones under special conditions.

Some well known interconnection networks are complete graph

,linear array,mesh,ring hypercube etc.

10
Contd..

Figure 3: Complete graph

Figure 4: Linear array

11
Contd..

Figure 5: Mesh

Figure 6: Hypercube 12
Star Graph
Definition

• In a star network, we begin with giving an integer n and the

integer set {1, 2, ... , n}. Each processor corresponds to a distinct
permutation of these n symbols. Therefore, the total number of
nodes in this star network is n!.
• A processor p is connected to n-1 neighbours which can be
obtained by interchanging the first symbol of p with the ith
symbol, 2 ≤ i ≤ n. We call these n - 1 connections dimensions.
• We denote this network by Sn or n-star.
• For example, if n = 4, and processor P1234 is connected with
P2134, P3214 and P4231 by two-way links.

13
Contd...

• The degree of Sn is (n-1)!

14
Contd...

• Sn can be decomposed into n sub-stars Sn-1 namely, Sn-1 (i), 1 ≤ i

≤ n where all the vertices have i for their last symbol in their
binary representation.
• Sn-1 (i) is an (n-1)-star defined on symbols in {1, 2, ... n}-{i}
• S4 in Figure contains four 3-stars, by fixing 1, 2, 3, 4 as their last
symbols.

15
Contd...

Figure 8: Star graph of dimension 4

16
Shortest path routing algorithm

The routing between two arbitrary node in n-star reduces to the

routing from an arbitrary node to the special node labeled with the
identity permutation Id = 123. . . n.

This reduction is made possible by the vertex symmetry of the star

graph. To reach Id from some other permutation P in the n-star, it
suffices to use repeatedly one of the following two rules until Id is
reached:

17
Contd..

Algorithm
1. if symbol 1 is in first position then exchange it with any symbol
not in its position, and
2. if symbol i (i ̸= 1) is in first position, then move it to position i.

Example
13425 → 31425 → 41325 → 21345 → 12345
With shortest path algorithm the diameter of n-star graph is
determined as 3(n−1)
2

18
Parellel paths

It is important to have node-disjoint paths (or parallel paths)

between two nodes in an interconnection network to speed up
transfers of large amounts of data and provide alternative routes in
cases of node failures.

Since the n-star is vertex symmetric we reduce this problem to

constructing parallel paths between an arbitrary node and the
special node labeled with the identity permutation Id.

19
Contd...

Calculating parellel paths in Star graph

For each i, 2 ≤ i ≤ k, we construct a path from P to Id along which the
misplaced symbols 2,3, . . , kc (where kc is number of misplaced
symbols), are corrected in the order i, i + 1,.. . , kc ,2 ,3,.. . , i - 1.

So, there are m steps in the construction of paths, each step

corresponds to the correction of one misplaced symbol.

We apply the following rules to correct a symbol k:

• if k is in first position then take it to position k

• if k is not in first position then first exchange k with the symbol
in first position, then take k to position k.

20
Contd...

Condition 1 :
If A and B are two nodes of an n-star with respective labels P1 and P2
having the same first symbol, then there are n - 1 parallel paths
between A and B. Furthermore, m of these paths are of minimum
distance equal to c + m and the other n - m - 1 paths are of length
c+m+2
where c is the number of cycles in permutation and m is the
number of misplaced symbols

21
Contd...

example
Let P1 = 132546 and P2 = Id = 123456 be the labels of two nodes in the
6-star. The P1 -cycles are C1 = (2,3) and C2 = (4,5),c = 2 and m = 4.
We show below four parallel paths of minimum distance, c + m = 6
and other one path with distance c+m+2=8, between P1 and P2 .
1 = 132546 → 231546 → 321546 → 123546 → 423516 → 523416 →
123456
2 = 132546 → 312546 → 213546 → 413526 → 513426 → 213456 →
123456
3 = 132546 → 432516 → 532416 → 132456 → 231456 → 321456 →
123456
4 = 132546 → 532146 → 432156 → 234156 → 324156 → 423156 →
123456
5 = 132546 → 632541 → 236541 → 326541 → 623541 → 423561 →
523461 → 623451 → 123456
22
Contd...

Condition 2 :
If A and B are two nodes of an n-star labeled with two permutations
P1 and P2 having different first symbols, then there are n-1 parallel
paths between A and B. These paths are grouped as follows:

a) c paths are of minimum length equal to c + m - 2

b) m - c - 1 paths are of length c + m, and
c) n - m paths are of length c + m + 2

23
Contd...

example
Let P1 = 43215 and P2 = Id = 12345 be the labels of two nodes in the
5-star. The P1 -cycles are C1 = (4,1) and C2 = (2,3), c = 2 and m = 4.
1 = 43215 → 13245 → 23145 → 32145 → 12345
2 = 43215 → 23415 → 32415 → 42315 → 12345
These are the parallel paths of minimum distance, c + m - 2 = 4

3 = 43215 → 34215 → 24315 → 14325 → 41325 → 21345 → 12345

This is the path of distance , c + m = 6

4 = 43215 → 53214 → 13254 → 43251 → 23451 → 32451 → 42351 →

52341 → 12345
This is the path of distance , c + m + 2 = 8
24
Fault tolerence

The fault diameter of an f- connected graph G is the maximum

diameter of any graph obtained from G by removing less than f
nodes.
Star graphs can tolerate upto (2n− 5) faulty nodes and the fault
diameter is increased only by 2 in the worst case in presence of
maximum number of faults

25
Arrangement Graph
Introduction

• The arrangement graph is a generalization of the star graph

topology.
• The major drawback of star graph is its scalability. A star
network has n! number of processors, resulting in a large gap
between n-star and (n + 1 )-star which has (n + 1)! processors.
Therefore, we may face the choice of either too many or too few
available processors when solving a particular problem by using
the star network
• Arrangement graph overcomes this restriction by giving another
parameter k to control the number of nodes in the toplology.

26
Contd...

• Its nodes correspond to the arrangements of k elements chosen

out of the n elements
• Its edges connect nodes which correspond to arrangements
differing in exactly one of their k positions
• We denote this by An,k

27
Definition

• An An,k graph, denoted is specified by two integers n and k,

where 1 ≤ k ≤ n.
• The node set of An,k is the set of all k-permutations of n, denoted
by < V > = { p1 ,p2 ...pk | pi ∈ < 1,2, ... , n > and pi ̸= pj for i ̸= j}.
• The neighbours of a node p = p1 ,p2 ...pk are defined as follows:
• pi p2 ...p1 ...pk through an edge of dimension i, where 2 ≤ i ≤ k (swap
p1 and pi )’ this kind of edges are referred to i-edges.
• xp2 ...pi ...pk through an edge of dimension 1, where x ∈ < 1,2, ... , n >
- { pi | 1 ≤ i ≤ k }, this kind of edges are referred to 1-edges.

28
Contd...

Figure 9: Arrangement graph A4,3

29
Contd...

n!
• The number of nodes in the An,k is (n−k)! .
• when k = n - 1, such An,n-1 is isomorphic to the Sn .
• It is a regular graph with degree k(n – k)
• It is vertex symmetric and for edge symmetry :

1. every 1-edge is edge-symmetric with any other 1-edge.

2. every i-edge is edge-symmetric with any other i-edge
• As a result of the node symmetric property, any node in An,k can
be mapped to the identity node ek = 12...k , which implies that
routing between two arbitrary nodes reduces to routing from an
arbitrary node to identity node e

30
Cycle structure for permutation

• A permutation of n symbols can be viewed as a set of cycles in a

way that each symbol’s desired position is occupied by the next
symbol. This cycle representation will be used in routing
algorithms.

For permutation p = p1 p2 ....pk ,

• The set [n] - {p1 ,p2 ,....pk } is the n - k external elements of p from
[n] which are not used in the permutation p.
• the set {p1 ,p2 ,....pk } is the k internal elements of p from [n] that
are used in the permutation p.

31
Contd...

example

Consider the node p = 74253 of the A8,5 arrangement graph.

The cycle representation for this node will be C1 = (2 , 4 , 5 , 3) and

C2 = (1 , 7).

Cycle C1 is an internal cycle since all its elements are internal, but
cycle C2 is an external one because of element 7 which is a foreign
element.

First cycle means that in this permutation position 2 is held by 4,

position 4 is held by 5, position 5 is held by 3, and position 3 is held
by 2. In second cycle position 1 is held by 7.
32
Routing and Shortest path

• As star graph, because of the vertex symmetry in arrangement

graphs, routing between any two nodes can be mapped to
routing from an arbitrary node p to the identity node.
• Routing can be done by first writing cycle representation
for node p and then correcting them one by one. Correcting a
cycle means placing each of its elements in its correct position.

33
Contd...

• External cycle can be corrected by first moving its non-foreign

external element to its correct position held by next element,
this procedure will be repeated until the last element, a foreign
element, is made external when it is replaced by the previous
element.
• Internal cycle requires to exchange the first element of the cycle
with an arbitrary external element, x. So, the first element is
now an external one and can be moved to its correct position
which is occupied by the next element of the cycle. We will
continue this procedure until the last element of the cycle is
taken to its correct position making x external element again.

34
Contd...

example

Following the above procedure, the external cycle C2 = (1 , 7) in A8,5 of

the node p = 74253 is corrected as bellow:

74253 → 14253

while the internal cycle C1 = (2 , 4 , 5 , 3) of the node p’ = 14253 is

corrected along the path:

14253 → 14653 → 12653 → 12643 → 12645 → 12345

3
• The diameter of An,k is 2k

35
Parellel paths

• In An,k , given two nodes which are not connected by an 1-edge,

then cycles formed with these two nodes with their
1-neighbours are disjoint from each other.
• Henceforth, using this I-neighbour cycle structure idea of An,k we
can find a new way to decompose the (n, k )-star graph into
different vertex-disjoint paths and cycles.
n!
• An,k can be decomposed into (n−k+1)! vertex-disjoint cycles of
length n-k+1.

36
Contd..

example

if n = 5 and k = 2, A5,2 can be decomposed into five (5!/(5 - 2 + I)! = 5)

vertex-disjoint paths of length 4:

path 1 : 21 ↔ 31 ↔ 41 ↔ 51
path2 : 12 ↔ 32 ↔ 42 ↔ 52
path3 : 13 ↔ 23 ↔ 43 ↔ 53
path4 : 14 ↔ 24 ↔ 34 ↔ 54
path5 : 15 ↔ 25 ↔ 35 ↔ 45

37
Fault tolerence

3
The fault diameter of An,k is at most 2k +4

38
Tables

Table 1: Comparison

Topology Size Degree Diameter

3(n−1)
Star Graph n! (n-1) 2
n1 3k
Arrangement Graph (n−k)! k(n-k) 2

39
Conclusion
Summary

• Introduced the arrangement graph, as a generalization of the

well-known star graph topology.
• The arrangement graph presents more flexibility than the star
graph in adjusting the major design parameters: numbers of
nodes, degree and diameter.
• The arrangement graph has been shown to be vertex and edge
symmetric, strongly hierarchical, with maximum
node-connectivity and a small fault diameter. Understood
shortest path routing algorithm for star and arrangement graph
• Computed Parellel paths for both star and arrangement graph.

40
References
References i

Interconnection Networks by Sheldon B.Akers,fellow,IEEE and

Balakrishnan Krishnamurthy,Senior member,IEEE Transactions
on Computers, VOL. 38, NO.4
On Some Properties of the Star Graph ,KE QIU Command and
Control Division, Defence Research Establishment Valcartier
(DREV), 2459 Pie XI, Courcelette, PQ Canada, G0A 1R0 SELIM G. AKL
,Department of Computing and Information Science, Queens
University, Kingston, Ontario, Canada, KTL 3N6
A Comparative Study of Topological Properties of Hypercubes
and Star Graphs Khaled Day and Anand Tripathi, IEEE
Transactions on parallel and distributed systems, VOL. 5, NO. I

41
References ii

Arrangement graphs: A class of generalized star graphs by

Khaled Day and Anand Tripathi Depart- ment of Computer
Science, Unilwsity of Minnesota, Minneapolis, MN 55455, USA

42
Thank You !

42