os

24
CHAPTER 2
Review of Multistage Interconnection Networks
This chapter provides an overview of various Multistage Interconnection
Networks. It also carries the review of existing literature on the subject that has been
undertaken to find the gaps in the research.
2.1 Introduction
Much of the literature available on the subject proposes many different networks
between the extremes of single bus and the cross bar to solve the problem of providing
fast, reliable and efficient communication at a reasonable cost in large parallel
processing systems. Such interconnection networks can be constructed from single or
multiple stages of switches. In a single stage network, data may have to be passed
through the switches several times before reaching the final destination. In multistage
network, one pass of multistage stages of switches is usually sufficient. The way input
units are connected with the output units, determine the functional characteristics of the
network, i.e., the allowable interconnections.
2.2 Construction and Performance of MINs
2.2.1 Cube Interconnection Networks
In a cube, vertical lines connect vertices whose addresses differ in most significant
bit position. Vertices at both ends of diagonal lines differ in middle bit position.
Horizontal lines differ in least significant bit positions. The unit cube concept can be
extended to an n-dimensional unit space, called n-cube, with n bits per vertex.

25
The implementation of a single stage cube network is given in Figure 2.1 for 8
nodes. The interconnection of the switching elements, corresponding to three routing
functions is shown separately in Figure 2.2.
Figure 2.1: A Three- dimensional Binary Cube
Figure 2.2: The Recirculating Cube Network for N = 8
The same set of cube routing functions, c0, c1, c2 can also be implemented by
using a three stage cube network. Two functions switch boxes, i.e., straight and
exchange are used in constructing multistage cube network. The stages are numbered as
0 at input end and increased to n-1 at output. The stage i implements ci routing function
6 72 3 4 50 1
6 72 3 4 50 1
6
6
u
72 3 4 50 1

26
for i = 0, 1, 2…(n-1). So, switch box at stage i connects an input line to output line that
differs from it only at ith
bit position as shown in Figure 2.3. The generalized cube
network is a multistage cube-type network topology that was introduced as a standard
for comparing network topologies (Siegel and McMillen, 1981). The Generalized Cube
Network of size 8x8 is exhibited in Figure 2.4.
Figure 2.3 : A Multistage Cube Network for N =8
Figure 2.4 : Generalized Cube Network with N = 8
a
b
c
d
e
f
g
h
i
j
k
l
2
3
0
1
3
4
5
6
7
0
1
2
3
3
4
5
36
7
3

27
2.2.1.1 Extra Stage Cube (ESC)
The Extra Stage Cube, a fault-tolerant multistage interconnection network, was
proposed by Adams and Siegel (1982) for use in large-scale parallel and distributed
supercomputer systems. It has all the interconnecting capabilities of the multistage
cube-type networks that have been proposed for many super systems. The ESC network
is derived from the Generalized Cube with one additional stage along with a number of
multiplexers and demultiplexers. The ESC is capable of operating in both SIMD and
MIMD environments. The extra stage provides an additional path from each source to
each destination. The ESC is a single fault-tolerant (Adams, et al., 1984). An ESC of
size 8x8 is shown in Figure 2.5.
Figure 2.5 : Extra Stage Cube Network of 8x8
2.2.2 Shuffle Exchange Network
Shuffle exchange network is based on two routing functions, shuffle and
exchange. A perfect shuffle of N = 8 is presented in Figure 2.6 (a). Perfect shuffle cuts
the deck into two halves from the center and then intermixes them evenly. Inverse

28
perfect shuffle does the opposite to restore the original ordering as shown in Figure 2.6
(b). These shuffle exchange functions can be implemented as either recirculating
network or a multistage network. Figure 2.7 produces a single stage recirculating
shuffle exchange network.
(a) (b)
Figure 2.6 : (a) Perfect Shuffle (b) Inverse Perfect Shuffle
Figure 2.7 : Shuffle Exchange Recirculating Network for N = 8
A multistage SEN has N = 2n
inputs, termed sources (S), and 2n
outputs, termed
destinations (D). It has n stages and each stage has N/2 switching elements. There is a
unique path between each source-destination pair. Figure 2.8 exhibits a SEN of size 8 x
8. The SEN is a self-routing network (Blake and Trivedi, 1988b). That is, a message
6 72 3 4 50 1
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7
0
1
2
3
4
5
6
7

29
from any source to a given destination is routed through the network according to the
binary representation of the destination's address. For example, if source S = 000 wants
to send a message to destination D = 101, the first bit of the destination address is used
by SE0,l for routing. So output link 1 of SE0,l is used for routing. At SE1,2 the second bit
of D is used and output link 0 of SE1,2 is used. Finally, at SE2,3 the third bit of D is used
and output link 1 of SE2,3 is selected as shown in Figure 2.9.
Figure 2.8 : Shuffle Exchange Network (SEN) of Size 8x8
Figure 2.9 : Routing in SEN for Communication Between S = 000 and D = 101

30
2.2.3 SEN with an Extra Stage (SEN+)
A SEN+ is an NxN SEN with an additional stage (Blake and Trivedi, 1988a).
The SEN+ system has N inputs and N outputs, with two paths between each source–
destination pair. It has n = (log2N)+1 stages and each stage has N/2 SEs. In general, the
switch complexity for the NxN SEN+ is N/2 ((log2N)+1). Thus, the additional cost of
SEN+ is N/2 switches or a fractional increase of 1/log2N, which is small for a large N.
Figure 2.10 demonstrates a SEN+ of size 8x8.
Figure 2.10 : SEN with an Extra Stage (SEN+) of Size 8x8
The addition of an extra stage to the SEN allows two paths for communication
between each source and destination (Gunawan, 2008b). A comparison of reliability
using the mean-time-to-failure by Blake and Trivedi (1988a) showed that SEN+ is
superior to SEN. Gunawan (2008b) compared three types of shuffle exchange network
(SEN) systems: SEN, SEN with an additional stage (SEN+), and SEN with two

31
additional stages (SEN+2). As measures of network performance, the terminal,
broadcast, and network reliability of these three networks were evaluated. It was
observed from the numerical results that the terminal reliability of SEN+ is the highest
among the three networks. Although the number of broadcast paths in SEN+2 is greater
than that of SEN+, the broadcast reliability of SEN+2 is the lowest among these three
networks. From this evaluation, it was concluded that SEN+ is the most reliable
network in terms of broadcast reliability. Similar conclusions were drawn for the
network reliability of the three systems. That is, SEN+ has the highest, while SEN+2
has the lowest reliability.
2.2.4 Augmented Shuffle Exchange Network (ASEN-2)
Augmented Shuffle Exchange Network (ASEN-2) is a regular network, having
equal number of switches in each stage. ASEN-2 network is constructed from Shuffle
Exchange network by adding 2×1 multiplexers at the initial stage and 1×2
demultiplexers at the last stage (Kumar and Reddy, 1987). It provides multiple paths
between any source and destination pair. The ASEN-2 shown in Figure 2.11 has a loop
size of two in every stage except the last one. Therefore, this network is called ASEN-2.
Switches that differ in the second most significant bit of their binary labels form the
loops in ASEN-2. In general, an ASEN in which the number of switches in a loop in ith
stage is equal to Min (2n-2-i
,K) is called ASEN-K. Blake and Trivedi (1988a) showed
that on the basis of reliability, ASEN-2 is superior to SEN and SEN+.

32
Figure 2.11 : Augmented Shuffle Exchange Network (ASEN-2) of Size 16x16
2.2.5 Omega Network
The shuffle exchange has been implemented with multistage Omega network.
Figure 2.12 presents an Omega network for N=8. An NxN Omega network consists of
log2N identical stages and between two stages there is a perfect shuffle interconnection
(Bhuyan, 1987). This network maintains a uniform connection pattern between stages.
Every input terminal has a unique path to every output terminal, and exhibits the
property of self-routing, i.e., routing is performed in a distributed manner using
destination address as the routing tag.

33
Figure 2.12: Omega Network of Size 8x8
2.2.6 Phi Network (PHN)
Sengupta and Bansal (2001) introduced a new fault-tolerant multistage
interconnection network named as Phi Network (PHN). PHN is an altered Omega
network retaining lesser number of switching elements (SEs) in the intermediary stages,
consisting of N/2 SEs of size 2x2 in the input and output stage, whereas the
intermediate stages consist of N/4 SEs of size 3x3. A Phi Network of size NxN, having
N sources and N destinations in n=log2N stage network, comprises of two groups,
denoted as Gi
. The input stage 1 has N/2 2x2 SEs, and intermediary (n-2) stages have
varying number of 3x3 SEs, interconnected through express links. Each of the N
sources is connected to both the groups through N 2x1 multiplexers before the input
stage and N 1x2 demultiplexers to the N destinations after the output stage. Stages are
numbered as 1,2, . . . (log2N), while switches in a group are numbered from top to

34
bottom chronologically as 0, 1, ... (N/4i
-1) for the ith stage. The express links have a
marked dominance on the fault-tolerance of PHN. Thus irregular PHN provides
different routes of differing path lengths providing much lower latency as compared to
the Omega Network. Figure 2.13 shows a PHN for N=8.
Figure 2.13 : Phi Network (PHN) of Size 8x8
2.2.7 Double Tree (DOT) Network
A DOT network is an irregular type of MIN. It consists of a right and a left half.
Each half of the network resembles a binary tree. The left and right trees are mirror
images of each other. A DOT network of size 2n
x2n
has 2n
input and 2n
output terminals
and 2n-1 number of stages (Mittal et al., 1995). Further, it has 2n+1
-3 switching elements
(Bansal et al., 1991b). Each of the stages i and 2n-i have exactly 2n-i
switching elements
of size 2x2 for i = 1,2,3,…..,n. A DOT network of size 8x8 is shown in Figure 2.14.
Kumar (2007) proposed a new mathematical model for performance evaluation of the
buffered Fault Tolerant Double tree packet switching network under uniform traffic

35
condition. An algorithm to implement this model is also presented and results obtained
are compared with the results of simulation. It has been shown that as the size of
network increases, throughput decreases and delay increases.
Figure 2.14 : Double Tree (DOT) Network of Size 8x8
2.2.8 Modified Double Tree Network (MDOT)
MDOT is an example of non-fault-tolerant irregular network, having different
number of switches at each stage. Figure 2.15 exhibits a modified double tree network
for N = 8. Total number of stages in this type of network are 2n-1, where n = log2N and
the number of switching elements are 2n+1
–3 (Aggarwal and Bansal, 2002). The number
of switching elements at the last and first stage is equal. Similarly, number of switching
elements in the next stages is equal.

36
Figure 2.15: Modified Double Tree (MDOT) Network of Size 8x8
2.2.9 Fault-tolerant Double Tree (FDOT) Network
An FDOT-k network of size NXN introduced by Bansal et al. (1991b) is
designed by dividing N inputs and N outputs into k disjoint partitions of N/k sources
and N/k destinations, where k >=2 and N >k, are the powers of 2. There are k
independent subnetworks, and an extra one, such that a connection path between
each source-destination pair can be established via any one of the subnetworks. All
the (k+1) subnetworks are of identical type. The extra subnetwork helps to enhance
fault-tolerant capability and to keep a desired level of performance even in the
presence of faults. Each source and destination is linked to all the (k+1)
subnetworks via kx1 multiplexers and 1xk demultiplexers. An FDOT-k consists of
(2n-1) number of stages and (k+1) (2n+1-3) number of switching elements, where n
= log2N/k. FDOT network of size 8x8 is shown in Figure 2.16. FDOT provides full
access capability in the presence of multiple faults. Bansal et al. (1991b) found that

37
FDOT supports on-line repairability and is cost-effective than other fault-tolerant
multistage interconnection networks, namely, INDRA, ESC and 3-Rep.
Figure 2.16 : Fault-tolerant Double Tree (FDOT) Network of Size 8x8
2.2.10 Extra Group Network (EGN)
Wei and Lee (1988) introduced a new class of fault-tolerant multistage
interconnection networks named as Extra Group Networks (EGNs). An EGN is
designed by dividing N sources and N destinations into several disjoint partitions. There
are several independent subnetworks (groups), one for each partition and the extra one,
such that a connection path of each source-destination pair can be established via any
one of the groups. The extra group helps to enhance fault-tolerant capability and to keep
a desirable level of performance even in the presence of faults. Thus, an EGN of size N
has one mxl multiplexer stage (input stage), log2(N/m) stages (intermediate stages) of
2x2 crossbar switches, and one 1xm demultiplexer stage (output stage) as shown in

38
Figure 2.17. There are N/2+N/2m switches in each of the intermediate stages, N+N/m
multiptexers in the input stage, and N+N/m demultiplexers in the output stage. Each
unique m path network of size N/m plus its associated multiplexers and demultiplexers
will be called a group. These groups including the extra one will be denoted by Go
, G1
, .
. . , Gm
. The EGN is more reliable and cost-effective than existing networks SEN and
INDRA.
Figure 2.17 : Extra Group Network (EGN) of Size 8x8
2.2.11 Augmented Baseline Network (ABN)
Bansal et al. (1991a) introduced a new class of fault-tolerant multistage
interconnection networks named as Augmented Baseline Networks (ABNs). An ABN is

39
a baseline network with two less stages, additional intra-stage auxiliary links,
multiplexers, demultiplexers, and slightly more complex switches. The ABN is more
reliable and cost-effective than baseline. Figure 2.18 shows an Augmented Baseline
Network (ABN) of size 16x16.
Figure 2.18 : Augmented Baseline Network (ABN) of Size 16x16
Liu et al. (2002) proposed a new type of MIN called Advanced Baseline Network
by using 4x4 switches instead of 2x2 switches. The Advanced Baseline Network has
simple routing technique and better performance but increased cost.
2.2.12 Four-Tree (FT) Network
An NxN FT is constructed by using two identical groups each consisting of
MDOT network of size N/2xN/2, which are arranged one above the other (Aggarwal

40
and Bansal, 2002). Both the stages i and 2m-i in the group have 2n-i-1
switches (where,
i =1, 2, 3….m, m= log2 N/2, and N= 2n
). The switches in a stage having the same
number are chained together to form a loop, except the last stage. Each source and
destination is connected with multiplexers and demultiplexers as shown in Figure 2.19.
Figure 2.19: Four-Tree (FT) Network of Size 16x16
2.2.13 Quad Tree (QT) Network
Bansal et al. (1992a) introduced a new fault-tolerant multistage interconnection
network named as Quad Tree (QT). An NxN QT is constructed by using two identical
groups each consisting of MDOT network of size N/2xN/2, which are arranged one
above the other as depicted in Figure 2.20. All the SEs are 3x3 in size except the ones in
the last stage which are 2x2. Each source and destination is connected to both the
groups through multiplexers and demultiplexers. The delay of the network is directly
proportional to the path length encountered on the way to the destination.

41
Figure 2.20 : Quad Tree (QT) Network of Size 16x16
2.2.14 Zeta Network (ZTN)
The idea of QT Network is used in the designing of a 16 x 16 multipath ZTN
(Nitin, 2006). ZTN has extra SEs in intermediate stages with additional express
chaining links to provide better fault-tolerance to the network. A ZTN of size 2n
x 2n
, n
= log2N, and m = log2(N/2) for 2n
sources and 2n
destinations, is constructed with the
help of two identical groups GN
[where, N = 0, 1], each consisting of a DOT network of
size 2n-1 x 2n-1, which are arranged one above the other. Figure 2.21 shows a ZTN of
size 16x16. The two groups are formed based on the most significant bit (MSB) of the
source-destination terminals. Each source and destination is connected to both groups
with the help of MUX and DEMUX. Nitin and Subramanian (2006) found that the
reliability and cost-effectiveness analysis of the hybrid ZTNs with ABNs and QTNs
shows a remarkable improvement in reliability.

42
Figure 2.21 : Zeta Network (ZTN) of Size 16x16
2.2.15 Gamma Interconnection Network (GIN)
A GIN of size N = 2n has n + 1 stages being labeled from 0 to n and each stage
involves N switches (Parker and Raghavendra, 1984). Basically, switches of sizes 1x3
and 3x1 are coupled with the first and last stage respectively. Moreover, each switch
located at intermediate stages is a 3x3 crossbar. And each switch number j at stage i has
three output links connecting to switches at stage (i + 1) according to the plus-minus-2i
function. In other words, the jth switch at stage i has three output links to switches [(j -
2i
) mod N], j, and [(j +2i
) mod N] at each consecutive stage. Figure 2.22 illustrates a
GIN network with size 8x8.

43
Figure 2.22 : Gamma Interconnection Network (GIN) of Size 8x8
Furthermore, to improve the fault-tolerant capability of GIN, several schemes
have been introduced, such as extra stage gamma network (Yoon and Hegazy, 1988),
CGIN (Chuang, 1996), composite banyan (Seo and Feng, 1995), PCGIN, FCGIN (Chen
et al., 2000) and B-network (Lee and Yoon, 1990). Among them, extra stage gamma,
CGIN, composite banyan network, FCGIN and PCGIN can provide at least 2-disjoint
paths. On the other hand, FCGIN and B-network use dynamic rerouting to tolerate
faults, but B-network cannot guarantee one-fault tolerant whereas FCGIN is one-fault
tolerant. Chen et al., (2003) proposed a new 3-disjoint path network namely, 3-disjoint
GIN (3DGIN). This 3DGIN can tolerate two links or switch faults, and its hardware
cost is almost equal to GIN.
2.3 Literature Review of Multistage Interconnection Networks
Many researchers have worked on Multistage Interconnection Networks. Some of
the most promising works have been presented in Table 2.1.

44
Table 2.1 : Literature Review of Multistage Interconnection Networks
S.
No.
Citation Main Contribution
1. Adams and
Siegel, 1984
In this paper, a survey of existing networks, namely, Extra
Stage Cube, Delta Network, Modified Baseline Network,
Data Manipulator and Gamma Network was made. Network
characteristics such as degree of fault-tolerance, routing and
permutation capability have been discussed.
2. Adams and
Siegel, 1982
This paper presented the Extra stage cube (ESC) Multistage
Interconnection network, derived from the Generalized Cube
by the addition of one extra stage of interchange boxes and a
bypass capability for two stages. It is shown that the ESC
provides fault-tolerance for any single failure.
3. Bansal et al.,
1994
This paper introduced a class of fault-tolerant MINs named
as Augmented Baseline Networks (ABNs). Performance and
reliability analysis has shown that ABNs are more reliable
and cost-effective than baseline.
4. Bansal et al.,
1991a
This paper introduced a new fault-tolerant Multistage
Interconnection Network named as Augmented Baseline
Network (ABN). Performance of ABN is analyzed and
compared with Omega and Crossbar networks.
5. Bansal et al.,
1991b
This paper presented a new fault-tolerant MIN named as
Fault-tolerant Double Tree (FDOT) network. Performance of
FDOT is analyzed and compared with ESC, INDRA and 3-
Rep networks. It has been shown that FDOT is more reliable

45
than ESC, and INDRA.
Table 2.1 (Contd.)
6. Bansal et al.,
1992a
This paper introduced a new fault-tolerant MIN named as
Quad Tree (QT). It has been shown that QT is single switch
fault-tolerant and is more cost-effective than ESC, INDRA and
ASEN-2 networks.
7. Bhuyan et al.,
1989
This paper described the basics of parallel systems,
classification of interconnection networks and basic
terminologies for performance evaluation of interconnection
networks.
8. Bhuyan, 1987 This article was a first attempt to survey fault-tolerant
networks. It compared the properties of different networks and
provided the reader with the state of the art in this area.
9. Blake and
Trivedi, 1988a
In this paper, authors compared the reliabilities of two fault-
tolerant multistage interconnection networks; one relies on an
extra stage for increased reliability, the other uses intra-stage
links. A comparison of the mean-time-to-failure and cost of
SEN+ and ASEN networks was presented, and it was shown
that on the basis of reliability, ASEN is superior to SEN and
SEN+.
10. Blake and
Trivedi, 1988b
In this paper, a transient reliability analysis of the SEN and
SEN+ networks has been performed. Exact closed-form
expressions for the reliability of 8x8 and 16x16 SEN and
SEN+ networks were derived. A comparison of the mean-

46
time-to-failure of these networks showed that on the basis of
reliability, the SEN+ is superior to the SEN.
11. Chen and
Chung, 2005
This paper presented a fault-tolerant network called CSMIN
with two disjoint paths between any source-destination pair to
guarantee one fault-tolerance. CSMIN has the ability to switch
packets between these two routing paths at each stage and
keeps low rerouting hops (one rerouting hop on average), the
collision ratio of CSMIN is lower than GIN, B-network and
CGIN. CSMIN has advantages over other related disjoint paths
networks (GIN, CGIN, B-network) in both re-routability and
fault-tolerance capability
12. Chen and Fu,
2004
This paper presented a minimal links traversed dynamic
rerouting network called MinGIN. MinGIN provided the
dynamic rerouting method to guarantee one fault-tolerance.
Moreover, MinGIN needs 0 rerouting hops to find an
alternative path. Because of the property of 0 rerouting hops,
MinGIN presented better fault-tolerance and throughput than
other fault-tolerant networks, viz. GIN, CGIN and B-networks.
13. Cheng and Ibe,
1992
This paper considered the reliability of extra-stage
interconnection networks. Three types of reliability, namely,
terminal, network and broadcast are analyzed.
14. Das et al., 1998 This paper presented an O(Nn) algorithm that checks whether
a given permutation P is admissible in an m stage SEN,
1≤m≤n, and determines in O(Nnlogn) time the minimum

47
number of stages m of shuffle-exchange required to realize P.
15. Das and Das,
1997
In this paper the restriction on the size of MIN has been
removed and a new NxN generalized shuffle-exchange (GSE)
with logN stages has been introduced. It has been shown that
the GSE is a full access MIN, which can interconnect any
number of processors at minimum cost. Routing in GSE is
simple.
16. Gupta and
Bansal, 2008
This paper provides fundamentals of interconnection networks.
A survey of static (Baseline Network, Shuffle Exchange
Network, Delta Network, Generalized Shuffle Exchange
Network, Data Manipulator Network, Indirect Binary n-Cube
Network and Butterfly Network) and dynamic multistage
interconnection networks(Alpha Network, Phi Network,
Augmented Shuffle Exchange Network, Augmented Baseline
Network and Quad-Tree Network) has been carried out.
17. Gunawan,2008a The reliability bounds methodology presented in this paper
provides reasonable and simple estimates for reliability of a
complex large-scale gamma network. The reliability block
diagrams of a gamma network are designed and utilized to
obtain a lower and upper bounds. An 8x8 gamma network
demonstrates the application and accuracy of the proposed
methodology.
18. Gunawan,
2008b
In this paper, three types of shuffle-exchange network (SEN)
systems are compared: SEN, SEN with an additional stage

48
(SEN+), and SEN with two additional stages (SEN+2). As
measures of network performance, the terminal, broadcast, and
network reliability of these three networks are evaluated.
19. Kumar and
Reddy, 1987
This paper described the structure, topology, redundancy graph
and reliability of SEN, ASEN, ASEN-2 and ASEN-MAX
MINs.
20. Kumar and
Reibman, 1989
To provide fault tolerance and improve system reliability and
performance, a class of fault-tolerant multistage
interconnection networks, called augmented shuffle-exchange
networks (ASEN) has been proposed. ASEN are gracefully
degradable; although an individual component failure reduces
ASEN performance, it does not cause a total network failure.
The paper analyzed how these component failures affect
ASEN performance.
21. Liu et al., 2002 On the basis of the idea of Delta and Baseline networks, a new
type of MIN, Advanced Baseline is proposed in this paper. The
new network is designed using 4 x 4 switches. This paper
expounds its topological properties, routing technique and
compares the performance of the Advanced Baseline with that
of the existing Crossbar, Delta and Baseline networks. It has
been shown that the performance/cost ratio of the Advanced
Baseline network is obviously superior to that of Crossbar,
Delta and Baseline networks.
22. Mahajan and
Vig, 2008
On the basis of the idea of Delta and Omega networks, a new

49
type of MIN, Advanced Omega is proposed in this paper. The
new network is designed using 4 x 4 switches. This paper
expounds its topological properties, routing technique and
compares the performance of the Advanced Omega with that
of the existing Omega Network.
23. Nitin and
Subramani-an,
2006
The reliability and cost-effectiveness analysis of the hybrid
Zeta, ABN and QT MINs has been made and compared in this
paper.
24. Nitin, 2006 This paper presented a new irregular MIN, viz. ZTN, which is
constructed from two identical DOT networks. The reliability
bound analysis of ZTN has been made and compared with the
existing ASEN-2 network.
25. Nitin et al.,
2009
This paper provides a more comprehensive and accurate
algorithm that always generates correct routing-tags for two
disjoint paths for every source-destination pair in the existing
CSMIN. The CSMIN has been modified to design new Fully
Chained CSMIN in order to reduce the cost.
26. Sengupta and
Bansal, 1999
In this paper, Fault-tolerant behavior of an irregular network,
Quad Tree has been analyzed and compared with a regular
network, Augmented Baseline Network, under fault-free
conditions and in the presence of faults.
27. Sengupta and
Bansal, 1998
Efficient and simple routing algorithms have been developed
for two Irregular MINs, viz. MFDOT and QT. Static routing
provides full access for MFDOT, whereas minimal multiple

50
paths to reach the destination are provided through dynamic
routing in QT network.
28. Sengupta and
Bansal, 2001
This paper introduced a new fault-tolerant MIN Phi. Further,
the fault-tolerance and performance abilities of the proposed
Phi Network are compared with its predecessor, Omega
Network.
29. Sengupta et al.,
2000
This paper studies the performance of a well-known regular
network, ASEN; irregular network FT; and the proposed new
irregular network, Smart Four Tree (SFT). It has been found
that the reliability of SFT network is better among these three
networks, at the expense of increased cost.
2.4 Gaps in Existing Research Literature
Most of the research has been carried out on regular networks. There is a lack of
research effort on Irregular MINs.
Comparative analysis of regular vs irregular MINs is either missing or non-
comprehensive.
For performance analysis, the exact design consideration has not been accounted
for. For example, all computations assume the switch size of 2x2.
2.5 Chapter Summary
In this chapter, different unique path and multipath Multistage Interconnection
Networks, possessing regular or irregular topology, have been examined. Diverse
techniques to provide fault-tolerance in MINs have also been investigated. These

51
investigations provide help in the design of new fault-tolerant, high speed switching
networks.
In the next chapter, the Design and Routing of three new fault-tolerant irregular
MINs, viz. IASEN, IFT and IABN have been proposed and compared with the existing
ones.

os

More Related Content

Similar to os (20)

Recently uploaded (20)

os