Design and Analysis of a Multicasting and Fault-Tolerant Optical
Crossconnect for All-Optical Networks
Chi-Yuan Chang, Sy-Yen Kuo*
Department of Electrical Engineering, National Taiwan University, Taipei, Taiwan E-mail: sykuo@cc.ee.ntu.edu.tw
chiyuan@lion.ee.ntu.edu.tw
Received March 31, 2003; Revised and Accepted June 5, 2003
Abstract. This paper proposes an optical crossconnect (MFOXC) architecture that supports multicasting and fault tolerance. First, a tap-based and two splitter-based MFOXC node architectures are presented for wavelength routed all-optical networks. Compared to the existing optical crossconnects, the proposed MFOXC node not only performs multicasting ef®ciently but also improves reliability signi®cantly. The MFOXC introduces a new feature (fault tolerance) and keeps the multicasting capability. It can be used at some critical points in a network to improve the overall reliability and multicast performance. Furthermore, the probability of maintaining fault free operations has been investigated for both MFOXC architectures. We present our evaluation results with a commonly used reliability measure, the mean time between failures (MTBF). Finally, we have proposed the cost and the sensitivity analysis for these MFOXC structures. The cost model and the sensitivity analysis show that the cost reduction in different components has various different impacts on the total cost of a MFOXC architecture. It can help us to know which component dominates the total cost and how to make a decision to choose among different MFOXC structures. The simulation results show that (1) the decrease of 75% in the cost of the N 6 N switch will result in the reduction of 20% in the total cost of the tap-based MFOXC, (2) the 1 6 2 switch has a big impact on the cost of the splitter-based MFOXC structures, and (3) the variation in the cost of the splitter does not introduce signi®cant disadvantage to the type II splitter-based MFOXC structure.
Keywords: wavelength routing, fault tolerance, multicast, all-optical networks, wavelength division multiplexing
1 Introduction
As the internet traf®c continues to increase exponen-tially, a wavelength division multiplexing (WDM) network with terabits per second per ®ber becomes a natural choice as backbone in the next generation optical internet. This results in the introduction of wavelength routed all-optical networks (WRAON) [1]. For a WRAON, circuit switching is preferred since the optical technology for implementing the intermediate node buffering, header recognition and processing, which are indispensable for packet switching networks, is not available yet [2±3]. In such a system, network failures could interrupt a large number of communication sessions in progress, such as voice and data. As a result, the design of a WRAON must incorporate mechanisms to protect against potential failures, such as node failures, link failures, channel failures, and optical switch failures.
On the other hand, multicasting (for one-to-many or many-to-many communications) is important and increasingly popular on the Internet (IP over WDM). For a WRAON, the optical crossconnect (OXC) plays an important role to realize switching. The OXC supports point to point connections which has been intensively investigated [4±6], and point to multipoint (multicasting) connections [7±11] because a lot of future broadband services are multicasting ones.
For WDMmulticast, a (optical) switch needs to have the light splitting capability in order to be able to multicast data in the optical domain. To realize optical multicasting, one can utilize optical power splitters. A power splitter is a passive device used to distribute the input signal to all outputs; thus providing multicasting in the optical domain without buffering. The inevitable power loss requires the deployment of ampli®ers to compensate for the splitting loss. In addition, cross-connects which are able to satisfy all
the different multicast demands must be equipped with a splitter for every wavelength on every input ®ber link.
Several OXC structures where proposed to support multicasting [7±11]. Five classes of multicasting OXC exist. First, a star coupler with inherent multicasting capability is used to construct a class of OXC [7]. But wavelength converters and tunable ®lters are required to work in a large wavelength range. Second, two-stage splitters and one-two-stage combiners are used to construct another class of OXCs [7±8]. Third, a splitter-combiner switch that has inherent multi-casting capability is employed to construct another class of OXC [9]. Fourth, a splitter-and-delivery switch (SaD) is proposed to build OXCs [10]. Fifth, a tap-and-continue switch (TaC) [11] is also built to form OXCs with zero power splitter. All these OXCs support multicasting. Note that switches with spilit-ting capability are usually more expensive to build than those without. Due to this reason, many researchers selectively let the switches in a network have splitting capability [12], which means that only a subset of the switches in a WDMnetwork supports light splitting. These researches all focus on designing an OXC with multicasting capability, but without fault tolerance.
The fault-tolerance is one of the most important measures in optical quality of service (QoS). Hence, in order to achieve protection against failures, spares must be provided for the corrupted traf®c to be restored. It is also desirable that these failures should endeavor to be handled within the optical network layer, rather than by higher layers. With the advent of WDMtechniques, it is possible to provide redundancy by means of spare wavelengths (channels) and switches. Several simple failure restoration techni-ques for WDMmesh networks have been proposed in Armitage et al. [13], Baroni et al. [14], Miyao and Saito [15], Crochat and Boudec [16], Caengem, et al. [17], Dighe et al. [18], Shiragaki and Saito [19]. These researches all focus on designing failure restoration techniques, but without taking the OXC structure into consideration. In addition, the impact on network reliability and cost of the development of the optical transport layer, optical networking, and optical networking elements are an active area of investiga-tion. The OXCs are envisioned as elements that can improve network reliability. The other parameters in terms of modularity, fanout capability, and power distribution are very important for designing the
network. Therefore, in this paper we will propose a series of multicasting and fault-tolerant optical cross-connect (MFOXC) architectures and compare them in terms of modularity, fanout capability, power dis-tribution, reliability, and cost sensitivity.
The rest of this paper is organized as follows. Section 2 introduces the basic architecture of the MFOXC. Two different types of MFOXCs will be presented. We also compare them in terms of modularity, fanout capability, and power distribution between the tap-based MFOXC and the splitter-based MFOXC. Section 3 describes the reliability model and requirements for analysis. Section 4 proposes the reliability evaluation and cost analysis for different MFOXC architectures. Section 5 concludes this paper. 2 Node Architecture
2.1 Architecture
Fig. 1 shows a WDMall-optical network employing wavelength routing, which consists of OXCs inter-connected by optical links. Each OXC node includes a workstation A; B; . . . ; E and an optical switch 1; 2; . . . ; 5. Each link is assumed to be bidirectional and consists of a pair of unidirectional physical links. An optical crossconnect is a device capable of routing a wavelength on an input link to any output link. However, two input links with the same wavelength cannot be routed simultaneously onto an output link, e.g., the link from node D to node C in Fig. 1. If there are m wavelengths on each link, the OXC may be viewed as consisting of m independent optical switches, one for each wavelength as shown in Fig. 2.
Each optical switch has N inputs and N outputs where N is the number of input/output links. There are no optical-to-electrical (O/E) and electrical-to-optical (E/ O) conversions, and hence no buffering is needed at the intermediate nodes in these all-optical networks.
The effect of wavelength converters on the signal quality bas been investigated by adding converters to the OXC. Wavelength converters are often desired in the OXC to make the network management much easier and reduce the blocking probability because of their signal regeneration and noise reduction capabil-ities [20±22]. Fig. 3 shows an OXC with wavelength converters. The drawbacks of the wavelength con-verter include not only the added higher cost and complexity to the system, but also the single fault problem. If the central optical switch is faulty, the whole OXC will fail if no redundancy is provided at the node level. In addition, the crosstalk problem can be serious if the number of wavelengths is large. In
contrast, the OXC without converters in Fig. 2 can tolerate a single fault in an optical switch. This is because any nonfaulty optical switch can replace a faulty optical switch and the resulting OXC can still be functioning with all the wavelengths except the wavelength used by the faulty optical switch. As for fault tolerance capability, an OXC without converters is superior to an OXC with converters. But for network performance (in terms of wavelength reuse), the reverse is true. For designing a highly reliable MFOXC in this paper, we adopt the MFOXC architecture without wavelength converters. In the following we will propose two different types of MFOXCs, tap-based and splitter-based.
2.1.1 Tap-based MFOXC
Fig. 4 shows an implementation of an N6N tap-based MFOXC. The tap-based MFOXC is an OXC with multicasting and fault tolerance capability, which uses wavelength-dependent optical switches (or mechan-ical switches). It uses a set of tap-and-continue modules (TCMs) on the right side of Fig. 4(a). In the TCM1 shown in Fig. 4(b), an extremely small fraction of the input signal Pin(e.g., (1/1000) Pin[23])
is tapped and forwarded to the local station. The remaining power of the order of 99.9% is switched to the other N 1 outputs. The tapping device used is fully programmable so that the tapped signal power is determined by the signal to noise ratio (SNR). To switch the signal to any of the N 1 outputs, Fig. 4(c) shows that the tapping devices (Tap) are used to implement the TCM4. A 16N TCMmodule has
log2N
d e stages where stage i has twice the number of taps compared to stage i 1. Fig. 4(d) shows an example of a TCM8. An input signal is tapped in using a tapping device. A small fraction of the signal power is directed toward the local station, while the rest continues to a multistage network of taps. By controlling the voltage on these taps, an input can be connected to any output port(s). Hence, the TCM8 module can support multicasting traf®c in the optical domain by controlling the voltage on these taps. For example, a multicasting session with destinations 1, 4 and 5 shown in Fig. 4(d) can be achieved by controlling the voltage on TCM1 and TCM4. The destinations 1, 4 and 5 have the signal power distribution 25%, 25% and 50%, respectively and all other nodes do not consume power. Therefore, this architecture saves extra power consumption when a node is not a destination.
Fig. 2. An OXC without wavelength converters.
In order to improve the fan-out capability of the multicast, the MFOXC needs to increase the level of TCMmodules. The power distribution is limited in our design. For example, the power distribution in
8(16) destinations is about 12.5% (6.25%) of the original source power. If the limitation of power distribution is exceeded, we need to add some extra optical ampli®er. The optical loss can be compensated
with optical ampli®ers (e.g., EDFA (erbium doped ®ber ampli®ers) or SOA (semiconductor optical ampli®ers)) properly distributed in the MFOXC (not shown in the ®gure).
On the other hand, the 261 SW element shown in Fig. 4(b) and the 361 SW element shown at the right side of Fig. 4(a) are used to select the alternative port of switch for normal operation, fault tolerance, or multicasting.
The TAPs shown at the left side of Fig. 4(a) are used to select switches in the MFOXC. A small fraction of the signal power is directed toward the fault tolerant switch, while the rest continues to a normal switch. The bene®t of a tap is that all the signal power is directed toward the normal switches in normal operation mode. Only little signal power is transferred to the fault tolerant switch. However, if the normal switch is faulty, by controlling the taps, the signal power is directed toward the fault tolerant switch. Because this behavior occurs in the optical domain, no other restoration in the higher level needs to be considered. It is also desirable that failures should endeavor to be handled within the optical network layer, rather than by higher layers.
In addition for routing and switching signals, the MFOXC also serves as a source and sink of traf®c in the network by an array of multi-wavelength transmitters and an array of multi-wavelength re-ceivers, respectively. The source (sink) is the start (end) of a connection. Each inbound link and outbound link has its associated receiver Rx and transmitter Tx, respectively. The bottom of Fig. 4(a) shows that each Tx or Rx is realized by an array of wavelength transmitters or an array of multi-wavelength receivers, respectively. Each optical switch is extended with one additional port to support inbound link and outbound link for multicast. The resulting MFOXC has 2N optical/mechanical switches with N 1 inputs and N 1 outputs as compared to N inputs and N outputs in Fig. 2.
2.1.2 Splitter-based MFOXC
The splitter-and-delivery (SaD) is a cross-connect with multicast capability that was proposed in Hu and Zeng [10]. A cross-connect consists of a set of SaD switches (see Fig. 5(a)) for each wavelength. A SaD switch consists of an interconnection of power splitters, optical gates (to reduce the excessive crosstalk), and photonic switches. Fig. 5(b) shows the organization of a cross-connect based on the SaD
switch. In addition to the SaD switches, demulti-plexers (multidemulti-plexers) are used to extract (combine) individual wavelengths. In the following we will propose two types of splitter-based MFOXC. Type I
Fig. 6 shows an implementation of an N6N splitter-based MFOXC. Its architecture is splitter-based on the SaD switch. In order to achieve robustness on the splitter for reliability, we duplicate the splitter on each SaD switch. In Fig. 6(a), a fault tolerant and multicasting (FTM) module is proposed. The input lightbeam is initially transferred to one of the branches by controlling the SE. Each branch is split into n branches and connects to a switch. Hence, any input of the splitter can be connected to any output branch.
Fig. 5. (a) An SaD switch, (b) an N6N optical cross-connect based on the SaD.
Fig. 6(b) is a switch module equipped with N FTM modules and 261 switches. Each branch is switch-able to an associated output by a 162 switch. Therefore, any input can be connected to none, one, several, or all the output ports. This features a multicasting capability.
To implement the FTMmodule, four types of components, i.e., two splitters, optical gates, 261 SW, and a 162 SE are integrated on a silicon board using planar silica waveguide technology [4±6]. The struc-ture shown in Fig. 6(b) not only has the advantage of integration for mass production, but also supports the modularity of splitter-based MFOXC.
Type II
Fig. 7 shows an alternative implementation of an N6N splitter-based MFOXC. Its architecture is very similar to that of tap-based MFOXC. The main difference between these two architectures is the splitter-to-n (SPn) module shown in Fig. 7(b) and the switch element (SE). The former is equipped with a 16N splitter to be as a light-splitter in order to
support multicasting in the optical domain. The latter is for selecting switches to operate for fault tolerance. Owing to the power spreading of light splitter, the optical loss is assumed to be compensated with optical ampli®er (e.g., EDFA or SOA) properly located in the MFOXC (not shown in the ®gure). The SE can be realized using a photonic directional coupler with electronic control [24]. By controlling the SEs and the switches, the input can be connected to any output port(s). Hence, the SPn module can support multi-casting traf®c in the optical domain by controlling these SEs and switches.
2.2 Further Discussion
In this section, we compare the multicast parameters in terms of modularity, fanout capability, and power distribution between the tap-based MFOXC and the splitter-based MFOXC.
2.2.1 Modularity and Expansion
It is desirable that an MFOXC has a modular structure and is expandable to allow new ®bers (®ber modular) and wavelengths (wavelength modular) to be added to keep pace with the future evolution for the multi-wavelength optical transport network. There are two kinds of expanding methods. One is to increase the number of ®bers and wavelengths only by a few. The expanding method is straightforward without chang-ing the layout of the original MFOXC. Large scale devices, e.g., an N N16 m m1 switch and an m m161 multiplexer are used as anticipated. The other is that the number of wavelengths and ®bers are increased signi®cantly. Take Fig. 8(a) for example. If N is expanded by K times, where K is an integer, the original MFOXC is consider as a main module (as shown in Fig. 4(a)). K main modules are connected by a common junction [26]. On the other hand, if the ®ber bandwidth is increased by K times, it is obvious that the channel spacing will become denser. The original multiplexers must be replaced. The remainder of the original MFOXC is unchanged and regarded as a main module. K main modules are connected together by the new multiplexers and 16K splitters, as shown in Fig. 8(b). All the expanding operations as described above do not reduce the multicasting capability of an MFOXC.
2.2.2 Multicast Fanout Expansion
Tap-based MFOXC: Fig. 4(c) shows the tapping devices (Tap) that are used to implement a TCM4 to
support multicast. An 16N TCMmodule has log2N
d e stages where stage i has twice the number of taps as stage i 1. Hence, an 16N6K TCM module has only logd 2NKe stages.
Splitter-based MFOXC type I: The expanding method is straightforward without changing the layout of the original architecture. A larger scale device switch and multiplexer are used.
Splitter-based MFOXC type II: Fig. 7(b) shows the splitting devices are used to implement the SPn module to support multicasting. An 16kN SPn module, shown in Fig. 9, consists of a 16k splitter and k 16N splitter modules.
2.2.3 Power Distribution
Power is one of the most important quantitative measures in optical networks. For successful deploy-ment of optical networks this measure needs to be considered at the design phase.
A. Tap-based MFOXC
Fig. 4(d) shows an example TCM8. An input signal is tapped in using a tapping device. A small fraction of the signal power is directed toward the local station, while the rest continues to a multistage network of taps. By controlling the voltage on these taps, an input can be connected to any output port(s). Hence, the TCM8 module can support multicasting traf®c in the
optical domain by controlling the voltage on these taps. For a multicasting session with k destinations k < N, it can be achieved by controlling the voltage on the TCM1 and TCM4. The k destinations have at least a 1=k signal power distribution and the other
nodes do not have any power consumption. For example, a multicast session with destinations 1, 4 and 5 has a signal power distribution 25%, 25% and 50%, respectively (fanout 8). Hence, this architecture can be easily cascaded without extra optical power ampli®ers by designing the network topology appro-priately.
B. Splitter-based MFOXC
The power splitter is a passive device used to distribute the input signal to all outputs. Hence, for a multicast request in MFOXC, all the outputs have the same power distribution. In other words, the power loss depends on the fanout of a splitter instead of a multicast group. For example, a multicast session with destinations 1, 4 and 5 has the same signal power distribution 12.5% (fanout 8). Compared with the tap-based MFOXC mentioned above, the cascading capability is poor.
3 Reliability Modeling 3.1 Reliability Requirements
The appropriate downtime allowance for an optical crossconnect is not easy to determine. One reason is the lack of a uniform de®nition of OXC. In general, component reliability values may come from ®eld data, data from vendors, estimates, and guesses. Field data and vendor data are generally acceptable for system reliability calculations. Because some of the switch technologies are new, estimates and guesses of the component reliability data are necessary for estimating the system reliability. The reliability data of the components that compose the OXCs were obtained from vendors [25]. The appropriate compo-nents and their FIT values are listed in Table 1.
We present ef®cient reliability evaluations for the OXC, tap-based MFOXC and splitter-based MFOXC, respectively. In our analysis, we assume that the components are defect-free initially. For the operating system there must be a method of detecting a failure and a technique for system recovery from faults. Furthermore, we assume that each component has an exponentially distributed lifetime (i.e., the failure rates for system components are time-independent) and the component failure duration is short relative to the time between failures, and that times between failures and the duration of the component failures are
Fig. 9. A 16kN splitter-to-kn module.
Fig. 8. (a) Expanding method to increase the number of ®bers by K times. (b) Expanding method to increase the number of wavelengths by K times.
independently distributed. Our calculations are based on failure rates given in Table 1.
The ®rst step in performing a reliability analysis is to obtain a de®nition of the system architecture from both the functionality and the reliability perspectives. A graphical representation of the architecture such as the reliability block diagram is very helpful in providing a precise and usable description of the reliability structure of the system. The reliability block diagram is based on the functionality of the system and is a method of representing the effects of all possible con®gurations of working and failed components on the functioning of the system. The reliability block diagram is obtained from the de®nition of the system failure. Each block in the diagram represents either a component or group of components that has two states: working or failed. 3.2 Reliability Function
The reliability function R t is the probability that a component (system) will survive until t (i.e., a failure occurs after t)
R t P X > t 1 F t
where X is a positive, continuous random variable, called lifetime, describing the duration time of failure-free operation and F t is a lifetime distribution. The reliability function can also be seen as a probability of failure-free operation during a speci®ed period of time 0 to t.
For a series system, the probability that the system operates at instant t is a product of the probabilities
that all of the components (units) are functioning at the instant t. Thus
R t Yn
i 1
Ri t:
To calculate the system reliability function, it is necessary to determine the reliability functions for all components. For exponentially distributed component lifetimes and component failure rates b1; . . . ; bn, the reliability function for a series system becomes
R t e b1 b2 bnt e bt:
For a parallel system, the probability that the system operates at instant t is the probability that at least one component operates at instant t. In the case of active redundancy, it is obtained that
R t 1 Yn
i 1
1 Ri t:
4 Reliability Evaluation and Cost Analysis 4.1 Optical Cross-Connect Architecture
We ®rst consider two optical cross-connect architec-tures, with and without wavelength converters, as illustrated in Figs. 2 and 3, respectively. We consider the connection of eight wavelength channels on one speci®c input link with one speci®c output link. A system failure occurs when one or more wavelength channels are lost due to the failure of system components.
Fig. 10 presents the reliability block diagrams derived for the OXC architectures shown in Figs. 2 and 3. The diagrams are based on the functionality of the OXC nodes and are obtained from our de®nition of system failure. In Fig. 10, the failure rate for each
Table 1. Failure rates.
Component Symbol Failure Rate
1: 2 splitter bS 50 FITs
162 switch bSE 50 FITs
Tap bTap 50 FITs
Tunable Tx bTT 745 FITs
Tunable Rx bTR 470 FITs
WDMcouple (8) bWDM 360 FITs
Transmitter bT 186 FITs
Receiver bR 70 FITs
Optical gate bgate 40 FITs
16616 switch matrix bOM 1000 FITs
Wavelength converter bConv 2000 FITs
8-wide MT connector bMT 200 FITs
1FIT 1 failure/109h
Fig. 10. Reliability block diagrams for (a) OXC without wavelength converters and (b) OXC with wavelength converters.
block is indicated (for description of the symbols and their values, see Table 1).
4.2 Tap-based MFOXC
Fig. 4 shows an implementation of an N6N tap-based MFOXC. The tap-based MFOXC is an OXC with multicasting and fault tolerance capability. Fig. 11(a) show the reliability block diagram derived from Ali and De [11]. Fig. 11(b) presents the reliability block diagram derived from our proposed MFOXC archi-tectures shown in Fig. 4. The diagrams are based on the functionality of the OXC nodes and are obtained from our de®nition of system failure. In Fig. 11, the failure rate for each block is indicated.
4.3 Splitter-based MFOXC
Fig. 6 shows an implementation of an N6N splitter-based MFOXC. Its architecture is splitter-based on the SaD
switch. Fig. 12 presents reliability block diagrams derived for the MFOXC architectures shown in Figs. 5, 6 and 7, respectivity. In Fig. 12, the failure rate for each block is also indicated.
4.4 Results and Discussion
There are a variety of measures and functions of reliability that can be derived using the reliability block diagram and the failure rates of components and taking into account maintenance principles. In this paper, we consider the following:
1. Mean time between failures (MTBF).
2. Reliability function, i.e., the probability of error-free operation until t.
The reliability function becomes important when one has to determine the probability of failure-free operation during a speci®ed period of time (e.g., 10 or 20 years). For example, network operators may change some equipment after 10 or 20 years of operation due to, for example, the development of new technology. In such cases, they may require a high probability of error-free operation of this equipment during this time period. Numerical results for the considered OXC nodes are shown in Table 2. The results presented in Table 2 show that, from a reliability point of view, (1) an OXC without converters is superior to an OXC with converters. (2) the MFOXC has better performance in terms of
Fig. 11. Reliability block diagrams for (a) tap-based OXC and (b) tap-based MFOXC.
MTBF and reliability, and (3) the splitter-based MFOXC has better performance than the tap-based under the same reliability requirement.
Fig. 13 shows the reliability functions in different OXC structures. It is of interest to note that the splitter-based MFOXC, tap-based MFOXC, and OXC without wavelength converter have better
perfor-mance in reliability than those without fault tolerance (FT). On the other hand, the tap-based MFOXC has almost the same reliability with the splitter-based type II. To sum up, we can conclude as follows: (1) the reliability of an OXC without wavelength converters is better than that with wavelength converters. In this case, wavelength converters could improve resource
Table 2. Reliability results for a 16616 OXC. OXC Without
Wavelength Converters
OXC With Wavelength
Converters Tap-BasedOXC [11] Tap-BasedMFOXC Splitter-BasedOXC [10]
Splitter-Based MFOXC TYPE I Splitter-Based MFOXC TYPE II MTBF (year) 12.4 5.7 12.4 19.97 18.8 22 20 MTBF (hours) 1:096105 0:56105 1:096105 1:756105 1:656105 1:936105 1:86105 R t 10 years 69% 25% 41% 71% 51% 72% 71% R t 20 years 40% 19% 23% 38% 25% 47% 41%
utilization (in term of wavelength reuse), but not for reliability. This is trivial due to the fact that the wavelength converter is a complicated component. From the reliability point of view, an OXC without converters is superior to an OXC with converters. (2) Compared to the traditional structure, the proposed MFOXC node not only has the advantage of multicast capability but also improves the reliability with fault tolerance capability. (3) The reliability of the splitter-based MFOXC is better than that of the tap-splitter-based.
Indeed, we investigated the in¯uence of capacity expansion on the reliability of the MFOXCs con-sidered here. Increasing the capacity will affect the reliability of MFOXC. Table 3 shows the evaluation results for a 32632 OXC. The FIT value for a 32632 switch matrix is estimated to be 3125 (from the reliability data received for 16616 matrices [25]).
Comparing Table 2 with Table 3, the MTBF of tap-based OXC in Table 2 is about three times higher than that of a two times larger OXC in Table 3. This is due to the FIT value for a 32632 OXC is about three times higher than that for a 16616 OXC. On the other hand, the degradation of reliability for splitter-based OXC seems not signi®cant, i.e., the capacity expansion does not signi®cantly degrade the relia-bility of the splitter-based MFOXC. As a consequence, from the reliability point of view, we
can ®nd out that the splitter-based MFOXCs are attractive and more robust for large switching nodes. 4.5 Cost Analysis
However, the implementation cost is still a dom-inating factor in optical crossconnect. Generally speaking, the cost of a system is proportional to that of individual component. The component cost depends on the process of fabrication. Since the more complicated the fabrication process is, the less the reliability (the larger the FIT value) will be. Hence, we assume the cost of a component is proportional to the FIT values.
Some of the key design results are presented in Table 4. The total costs are normalized and expressed relative to the conventional OXC architecture. The major cost components modeled in this analysis included MUX, DeMux, N6N switch, Splitter, Tap, 162 switch, optical gate, and MT connect. However, the cost of ®ber was not included. The cost of each component was derived by considering the current FITs to estimate the component cost assumptions used. From Table 4, we can observe that the splitter-based MFOXC has the lowest cost, and about 20% (11%) less than the tap-based MFOXC.
In Table 4, we ®nd the splitter-based (type II) only uses a 16N Splitter with respect to 2N in type I. Because the large number of splitters in a multicast cross-connect has the negative implications of dif®cult and expensive fabrication, the type II splitter-based MFOXC is a better choice than the type I.
Fig. 14 shows the relative cost component break-down based on the four major categories: standard OXC, Tap-based, based type I, and
splitter-Table 3. Reliability results for a 32632 OXC. Tap-Based
MFOXC Splitter-BasedMFOXC Type I Splitter-BasedMFOXC Type II
MTBF (year) 7.6 9.9 7.8
MTBF (hours) 0:676105 0:876105 0:696105
Table 4. Cost analysis results for a 16616 OXC with eight wavelengths. Standard
OXC Tap-BasedMFOXC Splitter-BasedMFOXC (I) Splitter-BasedMFOXC (II)
DeMux (360x) N N N N MUX (360x) N N N N Switch NXN (1000x) M 2m 0 2m Splitter 16N (550x) 0 0 2N 1 Tap (50x) 0 2N 1m 0 0 1X2 SW (50x) 0 2m 1N 2mN 2m 1N Optical Gate (40x) 0 0 mN N MT connect (200x) 2 4 4 4 Costs 19 920x 54 320x 47 840x 42 910x
based type II. All values in the ®gure are relative to the conventional standard OXC. As shown, the N6N switch costs account for from one-fourth to almost one-third of the total cost in the tap-based and the splitter-based type II OXC. Furthermore, the 162 switch costs account for about one-third of the total cost in all types of MFOXC. The tap costs account for about one-fourth of the total cost in the tap-based MFOXC.
Moreover, in order to keep pace with the future advancements, a detailed sensitivity analysis was carried out for the main cost components to evaluate how the relative costs of the different technologies would change if the cost assumptions were not correct. The cost sensitivity analysis can help us to know which component dominates the total cost and how to make a decision to choose among different MFOXC structures.
The N6N switches, tap, and 162 switch have a large impact on the cost of the tap-based MFOXC. Therefore, they are considered ®rst. As shown in Fig. 15, the decrease of 75% in cost for the component tap, N6N and 16N switch results in the decrease of the
total cost about 15%, 20% and 18%, respectively in the tap-based MFOXC. On the other hand, the increase of 75% in cost for the component tap, N6N and 16N switch results in the increase of the total cost of about 18%, 22% and 20%, respectively in the tap-based MFOXC. Hence, the N6N switch has the largest sensitivity in cost ( 20% and 22%).
A sensitivity analysis is also performed for the splitter-based MFOXC. In Fig. 14 the N6N switches, tap, and 162 switch have a large impact on the cost of the type II splitter-based MFOXC. On the other hand, the splitter and 162 switch dominates the cost of the type I splitter-based MFOXC. As shown in Fig. 16, the type I MFOXC is more sensitive to the change in the splitter cost than the type II MFOXC. This fact shows that the larger impact to the total cost that the splitter has on the type I structure. The variation in the cost of the splitter does not introduce signi®cant disadvantage to the type II splitter-based MFOXC structure. However, the component 162 switch has almost the same sensitivity to cost on both splitter-based MFOXC structures.
Fig. 15. Cost sensitivity analysis for the tap-based MFOXC.
5 Conclusions
We have proposed several multicasting and fault-tolerant optical crossconnect (MFOXC) architectures. First, a tap-based and two splitter-based MFOXC node architectures were presented for wavelength routed all-optical networks. The tap-based MFOXC is an OXC with multicasting and fault tolerance capability, and uses wavelength-dependent optical switches. It uses a set of tap-and-continue modules (TCMs). The bene®t of a tap is that all the signal power is directed toward the normal switches in normal operation mode. Only little signal power is transferred to the fault tolerant switch. However, if the normal switch is faulty, by controlling the taps, the signal power is directed toward the fault tolerant switch. Second, a splitter-based MFOXC architecture based on the SaD switch was proposed. In order to achieve robustness on the splitter for reliability, the splitter on each SaD switch is duplicated. An alternative implementation of an N6N splitter-based MFOXC is very similar to that of the tap-based MFOXC. The main difference between these two architectures is the splitter-to-n (SPn) module. Compared to the traditional optical crossconnect, the proposed MFOXC node not only has the multicasting capability but also improves its fault tolerance capability. It can be used in some critical points in a network to improve the reliability and multicasting performance.
We also proposed two different conditions for the modularity and multicast fanout expanding. They are arranged in the wavelength or ®ber modular layout. Therefore, the expansion is simple if the MFOXC is expanded by allowing only a few ®bers and wavelengths added. The expanding method is quite different if the number of ®bers and wavelengths is increased signi®cantly. The original MFOXC is regarded as a main module and a few main modules are connected together. All the expanding operations do not destroy the fault-tolerance property and multicasting capability.
We also performed the reliability evaluations for the OXC, the tap-based MFOXC, and the splitter-based MFOXC. We ®rst consider two optical cross-connect architectures with and without wavelength converters, respectively. The simulation results suggest that the reliability of an OXC without wavelength converters is better than that of an OXC with wavelength converters. We also evaluated the
reliability of the tap-based MFOXC and the splitter-based MFOXC. Compared with the tap-splitter-based/splitter- tap-based/splitter-based OXC without fault tolerance, the proposed MFOXC node not only has the advantage of multi-casting capability but also improves the reliability and the fault tolerance capability. From the reliability point of view, the splitter-based MFOXC has better performance than the tap-based under the same reliability requirement. Therefore, it is attractive and robust for large switching nodes.
Finally, the cost model and the sensitivity analysis results show that the cost reduction in different components would have different degrees of impact on the total cost of the MFOXC architectures. For the tap-based MFOXC, the total cost is sensitive to all the critical components variations in cost. The cost decrease of 75% in the component N6N switch will result in the decrease of the total cost of about 20% in the tap-based MFOXC. The 162 switch has a larger impact on the cost of the splitter-based MFOXC structures and the variations in the cost of the splitter do not cause signi®cant disadvantage to the type II splitter-based MFOXC structure.
Acknowledgment
This research was supported by the National Science Council, Taiwan, R.O.C. under Grant NSC 90-2213-E002-113.
References
[1] R. Ramaswami, K. N. Sivarajan, Routing and wavelength assignment in all-optical networks, IEEE/ACMTrans. on Networking, vol. 3, no. 5, (Oct. 1995), pp. 489±500. [2] P. E. Green, Optical networking update, IEEE Journal on Select.
Areas in Commun., vol. 14, no. 5, (June 1996), pp. 764±779. [3] S. Bannerjee, C. Chen, Design of wavelength-routed optical
networks for circuit switched traf®c, Proc. Globecom '96, (London, UK, Nov. 1996), vol. 1, pp. 306±310.
[4] Y. D. Jin, M. Kavehrad, Optical cross connect based on WDM and space-division multiplexing, IEEE Photon. Technol. Lett., vol. 7, no. 8, (Nov. 1995), pp. 1300±1303.
[5] A. Watanabe, O. Okamoto, M. Koga, K. Sato, M. Okuno, 8616 delivery and coupling switch board for 320 Gbit/s throughput optical path cross-connect system, Electron. Lett., vol. 33, no. 1, (Jan. 1996), pp. 67±68.
[6] S. Johansson, M. Lindblom, P. Granestrand, B. Lagerstrom, L. Thylen, Optical cross-connect system in broad-band networks: System concept and demonstrators description, IEEE/OSA Journal of Lightwave Technol., vol. 11, no. 5, (May/June 1993), pp. 688±694.
[7] S. Okamoto, A. Watanabe, K.-I. Sato, Optical path cross-connect node architectures for photonic transport network, IEEE/OSA Journal of Lightwave Technol., vol. 14, no. 6, (June 1996), pp. 1410±1422.
[8] E. Iannone, R. Sabella, Optical path technologies: A comparison among different cross-connect architectures, IEEE/OSA Journal of Lightwave Technol., vol. 14, no. 10, (Oct. 1996), pp. 2184±2196.
[9] Y. D. Jin, M. Kavehrad, Optical cross connect based on WDM and space-division multiplexing, IEEE/OSA Photon. Technol. Lett., vol. 7, no. 11, (Nov. 1995), pp. 1300±1303.
[10] W. S. Hu, Q. J. Zeng, Multicasting optical cross connects employing splitter-and-delivery switch, IEEE Photon. Technol. Lett., vol. 10, no. 7, (July 1998), pp. 970±972. [11] M. Ali, J. S. De, Cost-effective implementation of multicasting
in wavelength-routed networks, IEEE/OSA Journal of Light-wave Technol., vol. 18, no. 12, (Dec. 2000), pp. 1628±1638. [12] X. Zhang, J. Y. Wei, C. Qiao, Constrained multicast routing in
WDMnetworks with sparse light splitting, IEEE/OSA Journal of Lightwave Technol., vol. 18, no. 12, (Dec. 2000), pp. 1917± 1927.
[13] J. Armitage, et al., Design of a survivable WDMphotonic network, Proc. of IEEE INFOCOM' 97, (Dallas, TX, USA, April 1997), vol. 1, pp. 244±252.
[14] S. Baroni, et al., Link failure restoration in WDMoptical transport networks and the effect of wavelength conversion, OFC' 97 Technical Digest, (Dallas, TX, USA, Feb. 1997), pp. 123±124.
[15] Y. Miyao, H. Saito, Optimal design and evaluation of survivable WDMtransport networks, IEEE Journal on Select. Areas in Commun., vol. 16, no. 7, (Sept. 1998), pp. 1190±1198.
[16] O. Crochat, J. L. Boudec, Design protection for WDMoptical networks, IEEE Journal on Select. Areas in Commun., vol. 16, no. 7, (Sept. 1998), pp. 1158±1165.
[17] B. V. Caenegem, et al., Dimensioning of survivable WDM networks, IEEE Journal on Select. Areas in Commun., vol. 16, no. 7, (Sept. 1998), pp. 1146±1157.
[18] R. Dighe, Q. Ren, B. Sengupta, A link based alternative routing scheme for network restoration under failure, Proc. Globecom '95, (Singapore, Nov. 1995), vol. 3, pp. 2118±2123. [19] T. Shiragaki, H. Saito, Optimal protection architecture for reliable dense-WDMlightwave networks, Proc. ICC '98, (Atlanta, GA, USA, June 1998), vol. 1, pp. 181±186. [20] B. Mikkelsen, et al., All-optical noise reduction capability of
interferometric wavelength converters, Electron. Lett., vol. 32, no. 6, (1996), pp. 566±567.
[21] T. Gyselings, G. Morthier, R. Baets, Strong improvement in optical signal regeneration and noise reduction through asymmetric biasing of Mach-Zehnder interferometric all-optical wavelength converters, Proc. ECOC 1997, (Edinburgh, UK, Sept. 1997), pp. 188±191.
[22] W. V. Parys, B. Van Caenegem, B. Vandenberghe, Meshed wavelength-division multiplexed networks partially equipped with wavelength converters, Proc. OFC 1998, (San Jose, CA, USA, Feb. 1998), pp. 359±360.
[23] P. Prucnal, E. Harstead, S. Elby, Low-loss, High-impedance Integrated Fiber-Optic Tap, Opt. Eng., vol. 29, (Sept. 1990), pp. 1136±1142.
[24] R. A. Spanke, Architectures for guided-wave optical space switching systems, IEEE Commun. Mag., vol. 25, no. 5, (May 1987), pp. 42±48.
[25] L. Wosinska, L. Thylen, Large-capacity strictly nonblocking optical cross-connects based on microelectrooptomechanical systems (MEOMS) switch matrices: reliability performance analysis, IEEE/OSA Journal of Lightwave Technol., vol. 19, no. 8, (Aug. 2001), pp. 1065±1075.
[26] A. Watanabe, S. Okamoto, K-I. Sato, Optical path cross-connect system architecture suitable for larger scale expan-sion, IEEE/OSA J. Lightwave Technol., vol. 14, no. 10, (Oct. 1996), pp. 2162±2172.
Chi-Yuan Chang received his B.S. (1991) in Electronic Engineering form National Taiwan University of Science and Technology, Taipei, Taiwan, his M.S. (1993) in electronic engineering from National Central University, Tao-yuan, Taiwan. He is currently a doctoral candi-date at the Department of Electrical
Engineering, National Taiwan University, Taipei, Taiwan. His current research interests include fault-tolerant WDMnetworks and optical Internet.
Sy-Yen Kuo received his B.S. (1979) in Electrical Engineering from National Taiwan University, his M.S. (1982) in Electrical and Computer Engineering from the University of California at Santa Barbara, and his Ph.D. (1987) in Computer Science from the University of Illinois at Urbana-Champaign. Since 1991 he has been with National Taiwan University, where he is currently a professor and the
Chairman of Department of Electrical Engineering. He spent his sabbatical year as a visiting researcher at AT&T Labs-Research, New Jersey, from 1999 to 2000. He was the Chairman of the Department of Computer Science and Information Engineering, National Dong Hwa University, Taiwan from 1995 to 1998, a faculty member in the Department of Electrical and Computer Engineering at the University of Arizona from 1988 to 1991, and an engineer at Fairchild Semiconductor and Silvar-Lisco, both in California, from 1982 to 1984. In 1989, he also worked as a summer faculty fellow at Jet Propulsion Laboratory of California Institute of Technology. His current research interests include mobile com-puting and networks, dependable distributed systems, software reliability, and optical WDMnetworks.
Professor Kuo is an IEEE Fellow. He has published more than 180 papers in journals and conferences. He received the distinguished research award (1997±2003) from the National Science Council, Taiwan. He was also a recipient of the Best Paper Award in the 1996 International Symposium on Software Reliability Engineering, the Best Paper Award in the simulation and test category at the 1986 IEEE/ACMDesign Automation Conference(DAC), the National Science Foundation's Research Initiation Award in 1989, and the IEEE/ACMDesign Automation Scholarship in 1990 and 1991.
