View
7
Download
0
Category
Preview:
Citation preview
CHAPTER 5
PERFORMANCE ANALYSIS OF SWITCH ARCHITECTURE
The optical networks require large-scale optical switch configurations
or cross-connects in provisioning, controlling and managing the panes, which
are difficult to fabricate in a single chip Many optlcal switch matrix dev~ces
must he connected to attain large-scale opt~cal swtch network. Also the
existtng technology has no better solution for meetlng hlgh port count, low loss
challenges with less cross connects.
Clos (1952) introduced switch (three-module-stage) network is used
to Increase the number of ports. The resultant network based on the
conventional architectures suffers with heavy power loss due to relatively high
insertion loss of the individual optical switch matrix device. Spanke (1986)
introduced the two-module-stage expandable network (using tree architecture).
However, it has a problem of h ~ g h number of interconnections and modules.
To overcome this, a solution is proposed (Okayama 1998 and 2000)
by deriving a new type of two-module nethork fiom the three-module-stage
network shown in Figure 5.1.
It consists of modular architectures (a group of interconnected basic
susitch elements), namely. Cross bar, Benes, Spanke. and Spanke-Benes. So. the
performance of the large-scale sw~tch network can be determ~ned b! both -
switch elements and modular architectures From the general~zed three-stage
architecture, two types (Type I and Type 11) of low loss, expandable
architecture are derived (Okayama 1998 and 2000) and used for the
performance analysis. Their characteristics are compared with the exist~ng
structures such as crossbar and tree. The results obtained are found to be better
than the origlnal structures.
Spanke sw~tch fabric IS cons~dered for the modular based performance
analysis of the optical switch networks. In this modular architecture, in general,
the number of Inlets is equal to the number of outlets (symmetric type).
However, the two-module-stage n e t ~ o r k derlved from F~gure 5 l
(Okayama 2000) requires unequal number of inlets and outlets
(asymmetric type) in its modular level (though the overall large-scale switch
network is symmetrical). As of now. no model is available to charactertze the
parameters of the asymmetric modular configuration. Hence, a new generic
mathemat~cal model 1s proposed (Nakkeeran 2003a), which works for both
I16
symmetric and asymmetric Spanke modular configurations. In this model, the
important parameters of the switch fabric, namely, the number of switch
elements, number of crossovers, maximum loss and minimum loss are
formulated Us~ng them, the proper tit,^ of asymmetrlc Spanke optical s~hltch
fabr~c are analyzed and thelr behavioral curves are drawn. This model will be
helphl to design modular switch configurations (for both symmetric and
asymmetric types) with economical number of sw~tch elements and crossovers
and In turn improves the performance of large-scale sw~tch networks
The fundamental chara~ter~stlcs of the bas~c swltch element,
namely, sw~tchlng tlme, lnsertlon loss, ~solat~on, po\*er consumption, onioff
ratlo and polarlzat~on dependent loss also determines the performance of the
large-scale swtch networks Recently, large number of bas~c swtch elements IS
Introduced w ~ t h unproved characterlstic performance (M~dw~nter 1993 Fan
and Hooker 2000, Ruan 2001 and Papad~mltrlou 2003)
The following section presents the performance analysis of large-
scale switch networks. Subsequently, the mathematical modeling of asymmetrlc
Spanke switch fabric and the performance analysis of modular switch
architectures are outlined. Before conclud~ng the chapter, companson of
d~fferent elemental switches (based on sw~tching time and loss per basic swltch)
and a selection method to choose desirable type of structure (the 'structure' may
be large-scale switch network or modular switch fabric or basic switch element)
for the glven application are discussed.
5.2 PERFORMANCE AIVALI'SIS OF LARGE-SCALE SWITCH
NETWORKS
Two types of expandable two-module-stage networks with fewer
interconnections and modules are proposed b) Okayama (2000). The networks
described are der~ved from a gcneralrzed three-stage switch netaork
(Flgure 5 1) In the first type, the bullding blocks In each module are I xn and
nxm nonblocking switches (Okayama 1998). In the second type, a crossbar.
Banyan, or four-stage wide-sense nonblocking network 1s used as building
blocks The switch network proposed by Clos (former one) is named as Type-]
and the other (laner one) one IS cal!ed Type-I1 The interconnection pattern
d~ffers in these two architectures. To attaln a two-module-stage network m
Type 1, the three-stage network 1s divided at the middle stage r x r and each
part of ~t can be combined with first-stage or third-stage switches, The Type-I1
archltecture makes it easier to rearrange the network into two-module-stage due
to the segregation of interconnections. The mathematical models for the derived
two types of optical switch networks are also suggested by Okayama (2000).
Besides these two types, tree arch~tecture is considered to evaluate the
performance of large-scale switch network. Using the available model, the
number of switch elements reauired and the number of interconnections
118
resulted are studied in these switch networks. In addition, number of
interconnection results for various modular architecture of Type I1 is evaluated.
m W m PQ'
Figure 5.1 Generalized three-module-stage architecture
The performance characteristics of the Type-I and Type-I1
arch~tectures are compared with those of the Tree architecture and the results
are shown in Figures 5.2 and 5.3
I E+OO -- -- . - - -
16 12 M 128 256 512 1024 POR COWlt
Figure 5.2 Switch-elements characteristics
The number of interconnections and modules are lower for both
Type-I and Type-I1 than for the Tree type arch~tecture Total number of
~nterconnections and switch elements are also found to be lower than the Tree
ppe architecture. The Tqpe-I1 wide-sense non-blocking network exhibits the
least switch element count and number of interconnections for the given port
count.
The total number of interconnections varies with the type of modular
architecture (used in the case of Tbpe-I1 architecture) as shown in Figure 5.4
It is observed that there is less number of interconnections in the case of
Type-11 network when the switch configurations like Crossbar, Banyan or
Benes are used as modular architecture
] - - - -- -
16 32 M 128 256 1024 Port count
Figure 5.3 Number of interconnections
16 32 €4 128 256 512 1024 FDlt count
Figure 5.4 Type11 interconnections for different modular architectures
Symmetr~c Spanke modular arch~tecture is not a good candidate to
design large-scale optical swltch networks as 11 provides large number of
interconnections. However, it can be used in designing such large-scale sw~tch
network because of its asymmetrical model availability (Nakkeran 2002a).
Generally, the modular architectures of large-scale optical switch networks are
symmetncal. Moreover, asymmetric mathematical model for other modular
architectures are not available (except for Spanke arch~tecture). Hence. b)
introducing the asymmetric Spanke architecture, the number of switch elements
required and interconnections resulted (in designing large-scale switch network)
can be reduced relatively (detailed discuss~on is made in Section 5.3).
5.3 SPANKE ASYMMETRIC SWITCH CONFIGURATIOK
(AS MODULAR SWITCH F ZBRIC)
The Spanke swltch configurat~on is a popular configurat~on for
building large non-~ntegrated switches In a stnct-sense, 11 1s a non-blockmg
t\pe of arch~tecture The mathemat~cal model of Spanke switch fabnc for
scmmetr~cal type IS glven b) Ramasuami and S~carajan (2000)
Consider the Spanke switch given In Figure 5 5, there are four inlets
and three outlets (asymmetric type). This is usually designed by combining 4
number of 1x4 switch elements w ~ t h 4 number of 4x1 switch elements.
However, all the switch elements are not effect~vely utll~zed in this des~gn. B)
using the proposed model (Nakkeeran 2002b, 2003a) this 4x3 switch can'be
des~gned by combin~ng 4 number of 1 x 3 swltch elements (two I x2 In each
switch, the third output of the splitter is directly taken from the first stage
122
spl~ner output), along with 3 number of 4 x 1 (three 2 x 1 in each) switch
elements. This design requires relatively less number of switch elements
resulting less number of crossovers.
The parameters of this swltch can be estimated by the proposed
generic model. It works for both the s)mmeaic and the asymmetric
configuration, which is tabulated in Table 5.1
Figure 5.5 4x3 Spanke switch ( m a )
Table 5.1 Proposed generic model of Spanke switch fabric
Sl.no I Parameters I Proposed model definition
1
2
3
No of su~tches (S) 2mn-(m+n)
Crossovers (C)
Maximum loss (L,,) 1 r1o~~rnM1og2nl
4 M ~ n ~ m u m loss (L,,,) ! log2m~lo&nl
123
This model is applicable even if the inlets ('m') and outlets ('n') are
interchanged. It means, this can also be valid for the 3 x 4 switch fabric (men).
5.3.1 Proposed generic model
There are four parameters available with Spanke switch configuration
The mathematical formulation of these parameters 1s discussed by using the
following nomenclature.
m number of inlets
n number of outlets
Basic switch element 1s 2 x 1 'and I x 2
mxn interconnections.
m t l , increasing inlet by one
n+l, increasing outlet by one
S number of switch elements
C number of crossovers
maximum loss
& minimum loss
The possible relationships between m and n can be
I m > n (Concentration type of snitches)
i ~ . m<n (Expansion t p e of switches)
iii. m=n (Symmetric type of switches)
5.3.1.1 Number of switch elements
The Spanke architecture can be interpreted as 'q' back-to-back
connection of a tree structure of 'm' Inputs (Okayama 2000, and Ramaswarn1
and Sivarajan 2000). To obtain the number of switch elements in the given mxn
switch ma@ix, the number of splitters and number of combiners at the input and
output polnts respectively are to be calculated. There are 'm' number of I xn
spl~tters at the input polnt and 'n' number of mx 1 combiners at the output polnt.
Each of the I xn splitters conslsts of (n-1) sw~tches. Each of the mx l comb~ners
consists of (m-I) switches
Therefore
Total number of sw~tches at the Input polnt = m (n-I)
Total number of swctches at the output polnt = n (m-1)
Hence,
The total number of sw~tches nqu~red ) (Totai numM of jA /Total number of su~tches ( 10 ~ 4 1 2 ~ the ~ I Y C P m x n matrix rultches at nput ride 1 (at output ildc 'I I
= mln-I ) + n ( m - I 1 ( 5 1)
5.3.1.2 Number of crossovers
Crossovers result from the ~nterconnection between 'mn' mput polnts
and 'mn' output points of the glven mxn switch matrix (Lm and
Goldstein 2000, and Smyth 1988). There are 'm' number of lxn splitters and
'n' number of mx l combiners. Every splitter must he connected to each of the
combiners through a connection. The total splitter outputs ('mn' points) can be
grouped into 'm' groups each having 'n' points (indexed from 0 to nm-I)
Similarly, the total combiner inputs (mn points) can be grouped Into 'n' groups
each having 'm' points (indexed from 0 to mn-1) as shown in Figure 5.6. The i"
outlet of each splitter must be connected to the i* group of the combiner.
Splitters outlets inlets Combiners
FP., 0
Figure 5.6 Schematic diagram showing inlet and outlet groups for
crossover
One outlet of each splitter must be connected to each of the comb~ner
~nlet. Consider the first splitter, the veq first link i.e '0' (0" outlet of the
spl~tter) to '0' (0' inlet of the combiner) does not have a crossover. The second
link is between 1' outlet of the splitter and m" inlet of the second combmer. In
126
this case, there are 'm' preceding inlet points of the first combiner which might
have connection from each of the splitters already T'herefore, there are (m-1)
crossovers for the 1" to m' llnk For the third link between 2nd outlet and 2m"
inlet, there are rn-1 preceding inlet points for each of the two upper comb~ners.
Therefore, there are 2(m-1) numbers of crossovers. S~mllarly for the ( n - ~ ) ~ ~
link, between (n-1)' outlet and m(n-1)" inlet, there are n-1 preceding
combiners, each will have m-l number of links from all spliners Therefore.
there are (m-l)(n-I) number of crossovers.
Hence,
For the m' combiner the new crossover (crossovers accounted already
for the prevlous links) is zero.
'The new
The same logic can be extended as follo\ls to arrlve at a generic
'Number of already
I 4 crossover = (Mablx Inlet m) - (on the l~nk
made connection ~n Number of preced~ng each co~lbiner (lnclud~ng Ixr comb~ners ( lanes
present connection I <tromOto(n-1)) , i vanes from 1 tom) J
I
2n-I 1 (n-I )m+l 1 (m-2) x (n- I )
I1 Spl~tter Outlets of the Splitter I lnlets of the Combiner New Crossovers
I n
(m-l ) T _ ~ ~ l l t t e ~ - - .- -- - -- -- Outlets of the ~ ~ t i t t e r / Inlets o f thF~ombiner hew Crossovers
1 I (m-2) x 0
111 Spl~tter
I (m-2)n+ l m+(m- 1 , (m-(m-I)) x I (m-2)n+2 ! 2m-&(m- I ) I (m- (m- I )) x 2 ~ (m-(m-l))x3
I Outlets of the Splitter 2n 2n+l 2n+2
L i I ! (m-1)n-I (n- 1 )m+(m- I) 1 (m- (in-I)) x (n-I) -
I n+ l
lnlets of the Combiner , New Crossovers 2 I (m-3) 0 I
2m+l (m-3) x 1 2 m + 2 I (m-3) x 2
m+ l (m-2) x I
2n+3 ?m+3 I l
(m-3) x 3 --
! n+2 2m+l (m-2) x 2
7 n+3 3m-I (m-2) * 3 - -
128
The total number of crossovers through the links in the mxn matrlx
snitch is the summation of all crossovers.
m-1 n-l
Total number of crossovers = C x ( m - I)J ,.I I-"
It is known that
LHS of equation (5.4) = m(m - I ) / ?
LHS of equation (5 5) = n(n - I ) / 2
Substituting the equations (5.6) and (5 7) in (5.2) gives
Total number of crossovers = m(m -1)/2 x n(n - ] ) I2
= mn(m-l)(n - ] ) I 4 (5 8 )
5.3.1.3 Maximum and minimum loss
In Spanke symmetric switch architecture, the attractive propem 1s the
presence of uniform loss, i.e., maximum loss is equal to minimum loss or
129
otherwise all the paths from any inlet to any outlet have the same loss
(Ramaswami and Sivarajan 2000).
In the symmetric case, since it 1s a back-to-back connection of two
binary trees, its loss is calculated by adding log2n + log2n = 210g2n.
In the case of asymmetric architecture
Maximumloss = rlog, ~nl+r log , n l
Minimumloss =Llog2 m l + L ~ o ~ , nJ
where r 1 denotes that the fractional value is converted Into the next
h~gher numerlc value and L denotes that the value is truncated
5.3.2 Analysis of the proposed model
For a better understanding of these parameters with respect to the
asymmetric Spanke architecture, let us consider all possible permutations and
combinations of switch matrlx of orders less than or equal to that of given value
'n' i .e . , 1 x n, Ix (n-I), l x (n-2), . . . 1x1, 2xn, 2x(n-1). 2x(n-2), . .2xl,
and n xn, n x(n-I), n x(n-2). n x(n-3), . . .. n x l There are n2 total possible
comb~nations of switch architecture.
If the value of n = 256, and if all its possible combinations of switch
matrix orders are considered, the associated number of switch elements, number
130
of crossovers, maximum loss and minimum loss are calculated based on the
proposed model as follows.
5.3.2.1 Properties of switch elements
Analyzing the value of sw~tching elements (S), the follovving
properties are identified.
a) The number of switch elements of all possible combination of
matrix orders can be put m a mamx form (row index = inlets;
column ~ndex = outlets).
b) The matrix 1s found to be a symmetric matrix
C) It is observed that the Increase in inlets (m), outlets (n) or both
(m, n) results In the vanations of S as follou.
S(m,n) = 2mn - (m - n ) (5.11)
S(m,n t l )=(?m-l ) tS(m,n) (5.12)
S ( m t l , n ) = (2n-I)tS(m.n) ( 5 13)
The variation in the number of switch elements required for the given
outlets 'n', for the chosen inlets 'm'. is drawn as shown in the Figure 5.7.
I E r n - -
2 4 8 16 32 64 128 256
m (Inlets)
Figure 5.7 Number of switch elements
5.3.2.2 Properties of crossovers
The following properties are inferred from the values of the
crossover (C).
a) The number of crossovers of all possible comb~nation of matrix
orders can be put in a matrix form as done for switch elements.
b) This matrix is also found to be symmetrical.
C) The increase in inlets (m), outlets (n) or both (m. n) results In the
variation of C as follow
C(m,n)= mn(m-1Xn-1)14 (5.15)
C(m,n t l ) = mn(m-])I2 tC(m,n) (5.16)
The variation in the value of number of crossover for the given
outlets 'n', for the chosen inlets 'm', is shown in the Figure 5.8
Figure 5.8 Number of crossovers
5.3.2.3 Properties of maximum loss and minimum loss
Analyzing the value of L, and L,, the following properties are
found.
a) The amount of loss of all possible combination of matrix orders
can be put in a matrix fonn.
b) The matrix is symmetrical.
c) If 'm' is a power of 2, then maximum and minimum losses are
equal irrespective of whether it IS a symmetrical or an
asymmetrical architecture
L,, = L,,. lf m = 2 '
L,, t L,,,,,, , Otherwise
But MAX(L-)- MAX(L-)= 2 or 1
d) The histogram for the frequency of occurrence of loss reveals the
following find~ngs
i. For a given inlet 'm', among all the posslble cornblnational
architectures, the more frequently occurring (f) maximum
loss value is ml2 + I .
This is shown in the Figure 5.9
Loss vahk (lmear)
Figure 5.9 Frequency of maximum loss
ii. For a given inlet 'm', among all the possible combinational
architectures, the more frequently occurring (0 minimum loss
value is ( d 2 ) - 1 and for few cases it is (mi2).
rn m i.e., f ( ~ r N ( ~ m a x ( r n ) ) ) = - - l o r - 2 2
This is shown in F~gure 5.10
4 30 28 .BxB
010x10 5 2 5 - E
15
p t o
1 2 3 4 5 6 Loss Value (!]near)
Figure 5.10 Frequency of minimum loss
5.3.3 Results and discussion on the proposed model
The proposed model for the asymmetric optical suitch configuration's
properties and behavior are discussed (Nakkeeran 2003a) in the preceding two
sections. If inlet, outlet, or both are increased, the effects on the parameters of
the switch fabric properties are summarized in Table 5.2.
I. Increase m number of sw~tch elements for Increase In ~nlet or
outlet by 1 = (2m-1) or (2n-1).
135
ii. Increase in number of switch elements for increase in inlet and
outlet by 1 = 2(m+n).
iii. Increase in number of crossovers for Increase in inlet or
outlet by 1 = (mn/2)(m-1) or (mni2)(n-I)
iv. Increase in number of crossovers for increase in inlet and
outlet by 1 = (mn/2)(m+n).
Table 5.2 Increase in crossover and increase in switch elements for
increase in inlet, outlet or both of switch fabric
45
2 4 0 -
- --
i
z 0 -
0 2 4 6
-
8 10 12 Number of mlets or outkls
Figure 5.11 Difference in switch elements for increase in inlets or outlets
-- -
-+- Asynrneblc - Symrnemc .
Number of inkts or outlets
Figure 5.12 Difference in crossovers for increase in inlets or outlets
0 10 20 30 40 50 Number of swnch elements
Figure 5.13 Number of additional crossovers for increase in switch
elements
The requlred number of switch elements to realize the given mxn
asymmetric Spanke modular swtch IS less compared to the symmetric one
(Figure 5.1 I ) S~milarl!, the number of interconnections resulted in the case of
as!.mmetric Spanke modular switch IS also less (Figure 5 12). The result~ng
additional crossovers for increase In switch elements are shown in Figure 5.13,
which illustrates that the asymmetric switch fabnc is showing better
performance. Besides, the remaining asymmetric switch results for swltch
elements and crossovers slmply rollow the symmetric switch matrlx
characteristics.
138
5.4 PERFORMANCE ANALYSIS OF MODULAR SWITCH
FABRIC
The analysis made for Spanke switch fabric in the previous
Section (5.3) is extended to compare the performance of other modular switch
architectures, namely Crossbar, Benes, Banyan and Spanke-Benes. The
comparison IS made In terms of number of baslc switch elements required,
maximum loss and crossovers. The results are shown In the Figures 5.14, 5.15
and 5.16 respectively. The results of the maximum loss including the crossovers
(here, it IS assumed that the loss due to one crossover is equal to the loss
Introduced by one bas~c switch elemenl) are shown In Figure 5 17
For example, conslder an 'n' port switch based on any one of the
modular architectures. The number of bas~c switch elements required for nxn
port modular swltch will vary from one modular arch~tecture to another (since
the configurauon of modular s ~ ~ t c h I S unique). Consequentl!. the number of
crossover resulted and loss from Inlet ro outlet w ~ l l v a r y from one to another In
this analys~s, the value of 'n' IS chosen from 2 to 256 (i.e., 2x2 to 256x256
switch matrices) However, the higher order values (128 and 256) of 'n' are
hardly used In practical applications at modular level.
From the responses of the modular architectures, the consolidated
performance results of varlous modular switch fabrics are tabulated In the
Table 5.2
Table 5.3 Results of modular switch architectures
Though the Ban) an, Spanke-Benes. and Benes modular sw~tch fabr~c
exhibit low loss compared to Spanke and Crossbar, they cannot be used for
sw~tch fabric design if the port count is not in powers of two (inlet, m = 2") and
also for asymmetric cases Howeber, Spanke and Crossbar can be used in these
cases. Though S p d e possess loss un~formlt). ~t 1s not su~table for Integration
because of the large number of crossovers On the other hand, Crossbar has no
crossovers but the path loss varies widely. When crossover is considered,
Crossbar and Spanke-Benes have no crossovers (Figure 5.16). If loss due to
both crossovers and switch element is considered, Banyan exhibits the IOU
140
m m u m loss followed by Spanke-Benes and Benes (Figure 5.17). They also
have loss uniformity.
From the discussion. it is observed that no part~cular architecture
exhibits the best performance w ~ t h all characterist~cs Hence, a method is
introduced to select an optimum switching architecture for a required
applicat~on, which is discussed in Sect~on 5.6.
I E406 - - - +crossbar + k n s s
I E-05 +banyan + spanke + spanke-benes -- -- -- -
p l E-04
--
2 3 8 I6 32 64 128 256
Port count(n)
Figure 5.11 Snitch-count characteristics
-t crossbar -c benes t banyan - spanke + spanke-benes
2 4 8 16 32 64 128 256 Pari count@)
Figure 5.15 3laxirnurn loss characteristics
x spanke + benes
t banyan --
4 8 16 32 64 128 256
Pon count(n)
Figure 5.16 Crossover characteristics
-+-crossbar + spanke-benes + benes
I E+OO 4 8 16 32 64 128 2%
Poa, count(n)
Figure 5.17 Maximum loss including crossovers
5.5 S\I'ITCHING TECHNOLOGIES
S~milar to modular arch~tectwes the sw~tch~ng technolog~es also pla?
a stgn~ficant role tn determ~nlng the performance of the large-scale swltch
networks There are var~ous types of sw~tches avalable like mechan~cal,
elecuo-opt~c, thermo-opt~c all-opt~c MEMS bubbles acousto-optlcs etc
(Apawal 1989 M ~ d u ~ n t e r 1993 Hunsperger 1994 and P a p a d ~ m ~ a ~ o u 2003)
All these swltches can be classified based on the following cnterla
(Ramasuaml and S~varajan 2000)
i. Domain of processing of the signal (optical, electrical)
ii Princtple of operation (linear, non-linear)
,.. 111. Type of technologies (waveguide, free space, Opt~cal Integrated
Circuits (OIC))
iv. Nature of techniques (space, time, wavelength etc)
v. Application (provisioning, protection, packet switching, Optical
Add Drop Multiplexers (OADM) etc)
Moreo~er e\.en s ~ b ~ t c h general11 possesses the following
fundamental operating parameters (Hunsperger 1991) Based on these, a suitch
can be chosen for a panlcular need
I . On!off ratio or extinction ratio
l i Swrtching time ... 111. Insertion loss
IV Isolation loss
v Power consumption and
VI . Polarization dependent loss
In this section, the switching tlme and insert~on loss of commercially
ava~lable non-all-optical switches (8 h'pes) are compared and their results are
shown in the Figures 5.18 and 5.19.
Discussion
Electro-optic switches possess the least switching tlme It also
exhibits zero net-loss and low Polarization Dependent Loss (PDL) m the case of
144
Semiconductor Optical amplifier (SOA) switches (Fan and Hooker 2000)
Electro-holographic developments compete on scalability and are advantageous
when individual wavelengths need to be switched. Holographic switches also
possess nanosecond-sw~tching time and hence. 11 could be used for packet-by-
packet s\vitch~ng in opt~cal routers. 3D MEMS (ww\v.l~ghtreading.com 2000)
arrays are the right chocce for large-scale optical cross-connects, particularly if
groups of wavelengths are to be switched together from one fiber to another.
Liquid gratings have lot of potential for optical add-drop multiplexers because
they are able to sw~tch out a single wavelength from a group in a single
operation H o ~ e v e r . the! cannot comoete m~th MEMS when handling groups
of wavelengths
Tfle of swtch
Figure 5.18 Switching time characteristics of different switches
bubbles U s m o - mecharucal memr hqud acousto. holoprm electxo- optc pramgr OPbC optc
Type of rlvllch
Figure 5.19 Insertion loss characteristics
Figure 5.1 8 illustrates the switching time (in sec) of various optical
swltches Among the switches considered. the electro-opt~cal sw~tch possesses
fast switch~ng time (approximately 100ns).
Figure 5.19 shows the insertion loss (dB) of various optical switches.
Among the switches considered. the hologram switches posses low insertion
loss (0,OldBibasic element un~t ) In t h ~ s analysis, the Insertion loss of all the
sivitches 1s converted into loss ! basic element unlt (basic unit available of ~ t s
k ~ n d namely 2 x 2 or I x 2 switches).
From the results, it is observed that no particular switch element
exhibits the best performance with all charactenstics. Hence, a method is
Introduced for the select~on of best switch element for a required application,
which IS d~scussed in the following section.
146
5.6 SELECTION METHOD
A selection method is introduced to select a desirable 'structure'
('structure' may be large-scale architecturelmodular architecturebasic switch
element) from the available types. In this method, (prototype model) three types
of large-scale switch configurations, five Wpes of modular switch
configurations and around elght basic switch elements d~scussed In the preclous
Sections (5.2. 5.4 and 5.5) are considered. For each of the character~stics of the
structure, a sorted list (names only) 13 maintained based on the value for the
characteristics (in the des~rable order). The first 'k' (user input) percentile
entrles In the respectlve characteristic list of items are selected and printed in
appropriate format This process can be repeated for various applicat~onsluser
needs b! varying the value of 'k' The detalled procedure for selection of
particular type of structure from the ava~lable group is illustrated through the
flow chart as shown in Figure 5.20. The s)mbols used m this flowchart are as
follows
x 1 5 x 5 3 I-Swltch element. 2-Modular; 3-Netwok
n Number of ltems (for any value of x)
s[xl[ll Names of the ltems for each x
c[xl[ilbl Character~stic values for each item in each structure
Result[x][i]Li] Ordered copy of the item names (sorted based
characteristic value)
k Percentile value (for the requlred sorted list)
m Number of characteristic of the structure
Start
Read the no (n) of items in the ixl structure
D o t = l , n
Read the charactertsttc value C[X] [ I ]~ ]
Accord~ng to the order, copy the respect~ve I ttem names from s to ~esult[x][t]!~] /
Figure 5.20 Flow chart for the selection of required structure from the available set
Read the structure type (x)
Read k
1 D o j = l , m
1 Do I = 1 L'n
I D~splay the 11st for Result[x][~][l]
Figure 5.20 Flow chart for the selection of required structure from the available set (Contd.)
149
The appropriate value of 'k' can be found by trial and error to meet
the needs. When 'k' is 0.4 for the crossover characteristlc, Benes architecture
(which is in the positlon greater than 0.4) is rejected though it is in the second
position in the list for swltch count characteristic (Table 5.4). Spanke-Benes
architecture is also rejected due to the same reason for the switch count
characteristlc, although it is in the first pos~tion for the crossover characteristic.
To include these architectures in the preliminary selection 11sts 'k' value can be
p e n as 0 5 The change of 'k' value to 0 6 does not make an) difference in the
selected lists, but the maximum acceptable value Increases
5.7 RESULTS AND DISCUSSlON
From the performance anal!sis of large-scale sw~tch architecture. the
Type-] and Type-I1 large-scale architectures are proven to be more efficient
than the existing two-stage tree architecture. Interconnections and module
number are lesser for both Type-I and Type-I1 than those for the tree
architecture. As explained in Sub-section 5.3 and Table 5.1, the proposed
mathematical model for asymmeulc Spanke swltch configuration can fulfill the
requirement of large-scale s n ~ t c h network proposed by Okayama (1998 and
2000). The properties of this model are discussed. In addition, uslng the same
model, the symmetric Spanke switch configuration can also be analyzed.
Different modular architectures and switching devices are analyzed
using their mathematical model Comparison has also been made among them
150
From the results, it is inferred that a particular architecture (or device) which is
good with respect to a particular perfonnance parameter exhibits poor
performance for other parameter For e.g., LiNbO, electro-optic switch has the
fastest sw~tching speed but the insertion loss is higher than that of other
switches. To select the optimum architecture (or device) with overall
performance, a procedure is introduced. Using the selection method when 'k' is
chosen as 0.5, the selected list of modular architectures is shown in Table 5.4
Table 5.4 Selected list of ordered modular architectures
Banyan Crossbar Banyan
Spanke-Benes Banyan Benes
Banyan and Benes exhibit loss uniformity but in Spanke-Benes
the loss varles from n to n'2. Honeber. Banyan or Benes cannot be used for
asymmetric networks (~nput and output ports dlffer) and for port numbers.
uhich are not powers of two. Spanke and Crossbar can be used for such cases
Crossbar exhibits low loss and no crossovers but its insertion loss varies widely.
Whereas Spanke exhibits loss uniformity and it is also of strict-sense non-
blocking type, it still has d~fficulties in fabr~cation due to crossovers
For switch elements, when 0.4 percentile is requested the selected
lists are shown in Table 5.5.
Table 5.5 Selected list of switch elements
Elements with overall optimum perfoimance are electro-optic (SOA) switches
and the hologram. The electro-optic switch is the fastest switch available (in ns)
and the hologram has the least insertion loss (?dB for 240x240 switch).
However. the SOA electro-optic switch possesses zero net loss and is scalable
to thousands of ports. Electro-holographic developments compete on scalability
and have advantages when individual wavelengths need to be switched. Even
though the switching speed IS In terms of milliseconds, 3D MEMS arrays are
probably the right cholce for large-scale optical cross-connects
(Hiang ?003), particularly ~f groups of t~avelengths are to he switched together
from one fiber to another.
Switching Time
Electro-opt~c (LlNbOl)
Electro-opt~c (SOA)
Hologram
Insertion loss
Electro-opt~c (SOA)
Hologram
Bubbles
Acousto-optic I Llquld eratlnps
152
The optimum performance large-scale switch network is found to
be Tqpe-11 thinned Banyan network with Banyan as the module architecture
using SOA electro-optic swltch element.
5.8 CONCLUSION
The performance analys~s of the large-scale sw~tch network IS carr~ed
out for modular arch~tecture and d ~ ~ c r e t e ca~tch elements Type 11 thlmed
banqan network w ~ t h banyan as the modular arch~tecture using SOA electro-
opt~c sw~tch element IS found to be the optlmum set
The proposed mathematical model for asymmetric Spanke switch
fabric 1s analyzed The mathematical model also irorks hell for symmetric
Spanke sw~tch configuration
A selection method is proposed which compares the required
percentage of the properties among the available switching network, modular
configurat~on and discrete sw~tch element, and selects a suitable one for the
requ~red applicat~on.
Recommended