QoS-aware Wavelength Assignment with BER and ... - Semantic Scholar

2 downloads 37819 Views 168KB Size Report
As discussed in [1], deploying all-optical networks is promising, but also ... In an all-optical network, a data signal is ... BER blocking, or if the latency incurred in processing the call exceeds ... circuits for large file transfers or video-conferencing.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

QoS-aware Wavelength Assignment with BER and Latency Guarantees for Crosstalk Limited Networks Jun He and Ma¨ıt´e Brandt-Pearce

Suresh Subramaniam

Charles L. Brown Department of Electrical and Computer Engineering University of Virginia Charlottesville, Virginia 22904 Email: {jh2eu, mb-p}@virginia.edu

Department of Electrical and Computer Engineering George Washington University, Washington, DC 20052 Email: [email protected]

Abstract— Crosstalk originating from optical switching devices and demultiplexers can be the dominant physical impairment in large all-optical networks. When a centralized network controller must estimate the impact of crosstalk and other physical impairments on the quality of a lightpath, this can add significant latency. In this paper, QoS-aware wavelength assignment algorithms are proposed that consider both bit error rate and latency guarantees. A technique called wavelength ordering is presented to alleviate physical impairments and also decrease the processing delay due to bit-error-rate (BER) estimation. Simulations presented show that the wavelength ordering algorithm outperforms other wavelength assignments when both BER and response latency guarantees are enforced.

I. I NTRODUCTION As discussed in [1], deploying all-optical networks is promising, but also quite challenging as many novel problems must be anticipated. In an all-optical network, a data signal is conveyed by a lightpath (a route from the source to the destination consisting of one or more fiber links on a chosen wavelength) and encounters crosstalk from power leaking in optical crossconnects (OXCs) and imperfect WDM-demultiplexing. This additive interference becomes the dominating degradation as networks expand and wavelength density increases [2]. Crosstalk may cause the quality of the optical signal to degrade and become so poor that its bit error rate (BER) becomes unacceptably high. One challenge in designing alloptical networks is determining how to assign lightpaths to call requests such that physical-layer impairments and delay are contained. Wavelength assignment (WA) algorithms are needed to mitigate the impact of crosstalk and noise and provide quality of service (QoS) guarantees, so-called QoSaware algorithms [2]–[4]. Conventional studies on routing and wavelength assignment (RWA) have proposed algorithms for establishing lightpaths without considering any physical impairments [5]. A blocking event, called wavelength blocking, occurs when a lightpath cannot be set up due to shortage of a free route or wavelength1 . These techniques do not distinguish one wavelength over another based on QoS. In crosstalk limited 1 No wavelength conversion is considered as these devices are still in experimental phases of development.

networks, a call request can also be blocked in real practical networks if the chosen lightpath has unsatisfactory BER, called BER blocking, or if the latency incurred in processing the call exceeds a given timeout. In this paper the term quality-ofservice (QoS) blocking, denotes BER blocking and timeout combined, which depends significantly on the choice of WA algorithms. For instance, it is well-known that the wavelength blocking rate of random-pick (RP, choosing randomly amongst the available wavelengths) WA algorithm is worse than that of first-fit (FF, choosing the free wavelength with the lowest index) WA [5]. However, RP WA is a BER-friendly algorithm because it tends to geographically spread wavelength use across the network such that crosstalk effects are not likely to be severe [2]. Therefore, the overall blocking probability of RP WA may be better than that of FF WA [3], [6]. In the last few years, QoS-aware WA techniques have been the subject of intense research [2]–[4], [6]–[8]. Some techniques [2] only allow lightpaths to be established if the QoS requirement is met, thereby inducing a much higher blocking rate but assuring the quality of the lightpath. More sophisticated techniques select routes and/or wavelengths based on some criterion of performance, such as “BER-aware” [6], in which the algorithm chooses the best lightpath to use. In [7], QoS-aware adaptive routing and wavelength assignment (RWA) algorithms are proposed that incorporate BER information in RWA to yield better performance in terms of average BER and fairness among network users. In [8], the authors propose several methods to combat four-wave mixing (FWM) impairment. The existing QoS-aware RWA algorithms are inevitably more complex than their conventional counterparts due to exhaustive searches and BER estimation for each candidate lightpath, which is typically not only performed for each incoming request, but also for each lightpath that the proposed connection might disrupt. It is possible that a new lightpath has an acceptable QoS but provokes so much crosstalk in the network that the QoS for other lightpaths (called involved lightpaths) drops below threshold [6]. Moreover, the algorithm has to compute the BER on candidate lightpaths in real-time before accepting a call because of the presence or absence of other co-propagating lightpaths, i.e., the instantaneous network

1-4244-0353-7/07/$25.00 ©2007 IEEE

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

II. S YSTEM D ESCRIPTION In this section, we present our model and assumptions for calculating the latency and estimating the BER of the network. A. Setup Latency We consider networks with bidirectional links, each of which supports W wavelengths in each direction. The network is assumed to be managed by a centralized controller with an infinite queue of received calls as shown in Fig. 1. A call denotes a request for a lightpath from one node of the network to another. Calls are assumed to arrive at the network nodes according to a Poisson process with mean arrival rate λa (calls/unit time). Call durations follow an exponential distribution with mean value E[X] (unit time/call). Thus the offered load per node in the network is λa E[X] Erlangs, which is assumed to be equal for all nodes. Call k incurs a delay Da (k) during the call admission control (CAC) procedure. We assume a timeout mechanism with a user-adjustable delay bound Tmax , which in a particular network can be large or small depending on the application. Based on the additive property of Poisson processes, the total call arrival rate offered

to the controller’s queue is N λa , where N is the total number of network nodes. Request from Node 1

Reply to Node 1 Routing table

Request from Node 2

Reply to Node 2

Queue ......

Processor

......

......

state, affects the number of crosstalk terms, the saturation of amplifier gains, and the ASE noise in the EDFAs. Long delays can be induced by using these complex algorithms. In this paper, we focus on the problem of wavelength assignment as performed by a centralized network controller. To our knowledge, this is the first study to include both BER and latency thresholds when evaluating QoS-aware RWA algorithms. We consider latency as an important performance measure for QoS-aware RWA algorithms since we believe that the service providers have to set up lightpaths for the incoming calls in a very limited time. Dynamic restoration wherein backup paths are not reserved but are discovered dynamically when a failure happens is one situation. For example, the restoration time has to be less than 50 ms after a failure has been detected in SONET/SDH networks [9]. Latency-guaranteed applications are required in future networks with on-demand fast set-up of circuits for large file transfers or video-conferencing. Based on this more complete QoS requirement, we propose several WA algorithms and evaluate their performance through simulation experiments, as was originally done for BER-constrained WA in [2], [6]. One of the WA algorithms considered is a promising heuristic wavelength ordering algorithm originally described in [3]. The performance of the WA algorithms is measured by total call blocking probability (BP) with both BER and latency guarantees. The paper is organized as follows. Section II describes the network considered, including our models for calculating the latency and BER of the desired lightpath. In Section III, we present the QoS-aware algorithms to be compared, including a short overview of the wavelength ordering algorithm. We evaluate the performance of QoS-aware WA algorithms by simulation and show the advantage of wavelength ordering in Section IV. Conclusions are drawn in Section V.

Reply to Node N−1

Request from Node N−1 Request from Node N

BER estimator

Fig. 1.

Reply to Node N

Controller architecture.

The processing delay Dp includes the time required to check the routing table and the time to run the BER estimation. The time to estimate the BER of a call depends on network traffic, the network state, the lightpath lengths, the BER threshold, and the severity of the physical impairments. Queueing delay Dq , i.e. the time a request must wait in the queue, depends on the sum of the processing delay of requests before it in the queue. The time required for finding a candidate lightpath is denoted τLP ; the time needed for checking if the BER of a candidate lightpath is higher than the BER threshold is denoted τc ; the delay incurred in checking if the BER of another involved existing lightpath is higher than the BER threshold is called τo . Assuming there are C existing lightpaths, the delay in processing request k is calculated by the following equation if within the timeout threshold. Dp (k) =

m  i=1

(τLP + τc +

C 

τo I(k, i, j))

(1)

j=1

where m denotes the number of trials before the processor finds a “good” (BER less than threshold) lightpath or finishes checking all candidate lightpaths. I(k, i, j) is an indicator function given by  1 if LP (j) ↔ R(k, i); I(k, i, j) = 0 else. where LP (j) ↔ R(k, i) means lightpath j interferes with request k when wavelength i is chosen. The total latency for call k is estimated as the sum of the processing delay and queueing delay of request k if within the timeout threshold, and is defined by  Dp (k) + Dq (k) if within Tmax ; Da (k) = else. Tmax Transmission delay, propagation delay, and other delays are ignored in this paper. B. Optical Physical Impairments Model The optical impairments considered in this study are insertion losses of components, crosstalk, and noise, including amplified spontaneous emission (ASE) noise from the erbiumdoped fiber amplifiers (EDFAs), shot noise, and thermal noise at the receivers. The EDFAs are modeled as automatic gain controlled as in [6]. Fiber dispersion and nonlinearity, which

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

could be incorporated using models available in the literature, are not accounted for here. We use the crosstalk model described in [2], [6]. The interfering crosstalk can be either coherent or incoherent with respect to the transmitted signal and other crosstalks. For the analysis performed here, we assume that all crosstalk terms are incoherent. Crosstalk originates mainly from two components in the switching nodes: the switching fabric, in which a small fraction of a signal at the same wavelength is sent to the other output ports, and the demultiplexers, in which a part of power from other channels cannot be entirely filtered out. The crosstalk power due to demultiplexing leaks is channeldependent, which makes the crosstalk from adjacent channels more powerful than crosstalk from channels separated by one or more channels (non-adjacent-port crosstalk). In this paper, the crosstalk from the demultiplexers is referred to as adjacentport crosstalk; the power of non-adjacent-port demultiplexer crosstalk is modeled as exponentially decaying with spectral separation. The switching fabric crosstalk and adjacent-port crosstalk power level attenuations with respect to the main signal are denoted Xsw and Xadj , respectively. Assuming binary on-off keying (OOK) modulation, the BER can be approximated using the statistics of the received signal for both the “0” and “1” bits. Denote µ0 and σ0 the mean value and the standard deviation of samples for bit “0”, and µ1 and σ1 the mean value and the standard deviation of samples for bit “1”, respectively. Normally, both µ0 and σ02 are small and can be neglected. σ12 is the sum of variances from intersymbol interference (ISI), noises, and crosstalk components, which are all assumed to be independent [2]. Under a Gaussian noise assumption, the BER can be approximated by BER ≈ √ 0.5erf c(Q/ 2), where the Q factor can be written as [10] Q=

µ1 − µ0 . σ1 + σ0

When a lightpath is established or torn down, the BER for lightpaths that share one or more OXCs may change and must be updated to account for crosstalk components that are injected or removed. III. Q O S- AWARE WAVELENGTH A SSIGNMENT In this section, we first demonstrate how the QoS requirements are applied to WA algorithms by using the first fit (FF) and random pick (RP) algorithms. The wavelength ordering algorithm is presented and a new wavelength assignment algorithm, FF with wavelength ordering (FFwO) is proposed. We denote as QoS-aware WA algorithms that find lightpaths which exceed the threshold on BER and latency (for the new call and every already-established lightpath) by a successive process of trying available wavelengths. Algorithms not identified as QoS-aware test a single wavelength for QoS compliance (QoS-guaranteed) and then quit. The QoS-aware approach used here is similar to existing BER-aware WA algorithms [6] except that we have added a latency guarantee. Therefore, QoS blocking can be caused by excessive delay or BER of the lightpath; the network

provides the user both BER and latency guarantees. The flowchart in Fig. 2 illustrates the QoS-aware approach. The centralized controller services requests from the source node using a first-come, first-served (FCFS) policy. The size of the queue is assumed to be infinite. In this paper, we use static shortest path routing. A route is determined for the connection by consulting routing tables, which are created by routing protocols dynamically or statically. The request is processed by the CAC algorithm if the total delay is less than the maximum latency the network or users allow. The QoS-aware wavelength assignment algorithm checks whether the total delay exceeds Tmax after each iteration to avoid extra processing for a connection request which exhausts its delay budget after a few iterations of the wavelength search. The controller determines if available network resources (i.e., routes and wavelengths) are sufficient to meet the declared BER requirements of a connection request by using the BER estimator. If a viable lightpath is found within the allotted time, the call is accepted. Otherwise, the connection request is rejected. A request arrives

Dq < Tmax ?

Wait in the queue

Yes Routing

No

Block the request

No free wavelength No

Choose a new free wavelength based on wavelength assignment algorithm, FF and RP No

Accept the request

Yes

Fig. 2.

BER < threshold

Yes

Da < Tmax ?

BER estimation of the lightpath

Flow chart of QoS-aware WA algorithms.

Wavelength ordering was proposed in [3] as a QoSguaranteed technique to increase the chances that the one wavelength considered by the FF WA algorithm has a low number of adjacent-port crosstalk terms. To decrease the latency caused by BER estimation, a wavelength ordering technique can also be integrated in QoS-aware FF WA since it can increase the probability that the first wavelengths tried have sufficiently low BER, thereby decreasing the number m of BER estimation cycles needed in (1). The optimal wavelength ordering is dependent on the traffic pattern, making it a difficult problem. We describe a heuristic wavelength ordering algorithm to find a good ordering sequence independent of the traffic. The algorithm for ordering the wavelengths places first on the list wavelengths that are most separated in frequency. Wavelengths are added in an optimal order such that the frequency separation remains maximal. The separation is measured by the spectrum distance. If there are several wavelengths with the same crosstalk impact, the wavelength whose sum of crosstalk contributions to all already-sorted wavelengths is minimum is chosen. For example, if there are 8 wavelengths then the optimized ordering should be (1, 8, 4, 6, 2, 7, 3, 5). We compare three QoS-aware WAs: QoS-aware FF (QoSFF), QoS-aware RP (QoS-RP), and QoS-aware FF with wave-

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

length ordering (QoS-FFWO), all of which guarantee the lightpaths’ BER and maximum latency. In the following, we evaluate them in terms of blocking probability. IV. S IMULATIONS AND RESULTS We evaluate the QoS-aware algorithms’ performance on a scaled version of the NSF network depicted in Fig. 3, with 32 wavelengths per link in each direction. The number of calls generated is more than 5 × 106 in each simulation. We model the arrival process as Poisson distributed and the service time as exponentially distributed with E[X] = 100 seconds. Fixed shortest-path routing is assumed for all techniques. The BER requirement is a BER< 10−12 , as computed using the technique mentioned in Section II-B. The latency is estimated by Section II-A, with Tmax (also called timeout threshold) set to 20 milliseconds. Other parameters of the network used in the simulations are listed in Table I. No wavelength conversion is included, and thus a call has to use the same wavelength from source to destination. The results from the simulation can be scaled to real networks; we choose values that demonstrate the impact of latency and BER in terms of blocking probability. The vertical dashed line in the figures is used to indicate a common point in the different simulations. Six WA algorithms are simulated. Three QoS-aware algorithms described in Section III are compared to their QoS-guaranteed counterparts, denoted simply FF, RP, and FFwO, which block unsatisfactory lightpaths chosen by the corresponding WAs yet do not repeatedly try other possible lightpaths. They all provide a QoS guarantee for every communicating lightpath. For both QoS-guaranteed and QoS-aware WA techniques, the blocking probability measured includes calls blocked due to wavelength continuity constraint and the failure of the chosen wavelength to satisfy the QoS (BER or latency) requirement. 60 MI

WA

300

NY

75

120

300 75

60

CA1

120

60 CO

60

120

CA2

NJ

NE

UT 60

75

120

IL

60 120

PA

105

MD

150 300 TX 150

in units of 10km

GA

Fig. 3. Topology of a downsized version of the NSF network with 14 nodes and 21 bidirectional links. The number on the links represents the length of the links in units of 10km.

The average blocking probability (BP) for various network loads for each WA algorithm is presented in Fig. 4 for Xadj = −25 dB and Xsw = −45 dB. The BP caused by latency > Tmax and the BP caused by BER blocking for various network loads for each WA algorithm are presented in Fig. 5 and Fig. 6, respectively. In Fig. 4, QoS-FFwO can be seen

TABLE I N ETWORK S IMULATION PARAMETERS Parameters Number of Wavelength Wavelength spacing Data rate per channel Fiber loss (Lf ) ASE factor (nsp ) Laser source power τLP τc τo Tmax

Value 32 50 GHz 10 Gps 0.2 dB/km 1.5 0 dBm (1 mW) 100 µs 700 µs 700 µs 20 ms

to perform the best out of the six WAs, yielding the lowest average BP. QoS-FF is second-best in our simulation scenario. Note that in low traffic cases, the wavelength blocking is negligible and BP is dominated by timeout and BER blocking. Fig. 5, where the BP of QoS-FF from excessive latency is bigger than 10−4 in low traffic shows that the total BP of QoSFF is bounded by the BP from timeout; Fig. 6 shows that QoSRP is bounded by the BP from BER blocking. When traffic load increases, the wavelength blocking and BER blocking become dominant. Thus the BP curve of QoS-FF stays below that of QoS-RP. Note that QoS-FF is always worse than QoS-FFwO since the wavelength ordering technique decreases the adjacent-port crosstalk impairment and does not produce longer delays. QoS-RP has the worst performance among the QoS-aware algorithms when traffic is medium or high, notably because it has the most severe wavelength blocking due to the random wavelength usage [5]. As we expect, the performance of the three QoS-aware WAs is superior to the performance of QoS-guaranteed WAs. Yet, the QoS-guaranteed WAs have a faster response compared to their corresponding QoS-aware WAs, as seen in Fig. 5, where RP, FF, and FFwO have negligible BP due to timeout, too low to appear in the graph. QoS-FF causes the most BP due to latency. For QoS-RP, randomness in WA tends to spread the calls among the wavelength spectrum so that the probability that the chosen wavelength meets the crosstalk from other inbandwavelengths is smaller than that of QoS-FF. C Thus the term of j=1 τo I(k, i, j) in (1) for QoS-RP becomes much smaller than that for QoS-FF. The wavelength ordering technique forces the probability that the chosen wavelength meets demultiplexers crosstalk to be small, but it cannot decrease the switching fabric crosstalk. Therefore, QoS-FFwO still brings a bigger delay than QoS-RP so that the BP of timeout for QoS-FFwO becomes slightly larger than that for QoS-RP in high traffic, yet still smaller than that of QoS-FF, as shown in Fig. 5. In Fig. 6, the BP caused by physical impairments is reported. QoS-FFwO and QoS-FF have the best performance, followed by QoS-RP. The reason that QoS-FFwO is slightly better than QoS-FF in medium and high traffic situations is the same as why FFwO is better than FF: the ordering technique alleviates crosstalk from the demultiplexers. This advantage also helps it keep a better performance than QoS-RP. Note that

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

0

10

−1

−2

10

−3

10

FF RP FFwO QoS−RP QoS−FF QoS−FFwO

−4

10

−5

10

90

95

100

105 110 Load, erlangs

115

120

0

10

Blocking probability caused by latency > 20ms

10

10

10

0

−1

−2

−3

QoS−RP QoS−FF QoS−FFwO

10

FF RP FFwO QoS−RP QoS−FF QoS−FFwO

−4

−5

125

Fig. 4. Call blocking probability for the six WA algorithms with BER and latency guarantees.

−1

10

10

90

95

100

105 110 Load, erlangs

115

120

125

Fig. 6. Call blocking probability caused by unsatisfactory BER for the six WA algorithms with BER and latency guarantees.

lightpaths than QoS-RP. Thus QoS-FF outperforms QoS-RP in Fig. 6. Overall, QoS-FFwO provides the best performance measured by total average BP among the six WAs studied in all traffic loads.

−2

10

10

0

−3

10

10

−1

−4

10

−5

10

90

95

100

105 110 Load, erlangs

115

120

125

Fig. 5. Call blocking probability caused by latency bigger than Tmax for the six WA algorithms with BER and latency guarantees.

Blocking probability

Blocking probability

10

Blocking probability caused by unsatisfactory BER

10

10

10

10

10

QoS-FF may have lower BP from BER blocking than QoSFFwO in light traffic situations, because QoS-FF experiences more total blocking. In other words, QoS-FF blocks more calls thereby decreasing the actual traffic in the network so that it experiences less physical impairments than QoS-FFwO. FFwO is better than RP and FF has the worst performance among them. The reason that FFwO is better than FF is from the fact that the ordering technique alleviates crosstalk from the demultiplexers. As mentioned in Section I, the crosstalk level of RP is not likely to be as severe as FF because of the behavior of spreading wavelength use across the network. Thus the BP curve due to unsatisfactory BER for FF is higher than that for RP. But this advantage disappears when QoSawareness is applied to the WA algorithms because QoS-aware algorithms try other available lightpaths before blocking the calls. QoS-FF tries to pack connections into fewer wavelengths so that it is more possible that QoS-FF has more candidate

−2

−3

FF RP FFwO QoS−RP QoS−FF QoS−FFwO

−4

−5

−55

−50

−45

−40

−35

−30

Xsw, dB

Fig. 7. Call blocking probability for six WA algorithms with BER and latency guarantees.

We now evaluate the six WA algorithms for a fixed traffic load of 100 Erlangs and adjacent-port crosstalk Xadj = −25 dB as the switching fabric crosstalk level Xsw varies from −55 dB to −30 dB. QoS-FFwO has the best performance as seen in Fig. 7. QoS-FF is the second-best algorithm. As the level of crosstalk increases, QoS-aware WA algorithms need more time to obtain a “good” lightpath. The gap in BP due to timeout between QoS-FF and QoS-FFwO is decreased. At low levels of crosstalk, wavelength ordering helps QoS-FFwO setup the lightpaths rapidly. In the presence of strong crosstalk, physical impairments from switching fabric crosstalk force QoS-FFwO to iterate more times to find a workable lightpath so that QoS-

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the ICC 2007 proceedings.

V. C ONCLUSION

0

Blocking probability caused by latency > 20ms

10

QoS−RP QoS−FF QoS−FFwO

−1

10

−2

10

−3

10

−4

10

−5

10

−55

−50

−45

−40

−35

−30

Xsw, dB

Fig. 8. Call blocking probability caused by latency bigger than Tmax for six WA algorithms with BER and latency guarantees.

0

Blocking probability caused by unsatisfactory BER

10

In this paper, we study the impact of QoS incorporating both BER and latency guarantees. We describe a new heuristic wavelength ordering algorithm to design the channel frequency allocation in order to minimize the crosstalk due to adjacent channel power leaking. It was shown that the wavelength ordering technique not only alleviates physical impairments, but also decreases the latency. This advantage is obtained without any extra run-time computational expense. Because of this ordering technique, the QoS-FFwO performs better than other QoS-aware WAs in all cases. QoS-RP induces less delay than QoS-FF and QoS-FFwO in many circumstances. FFwO is the best QoS-guaranteed WA in our simulation. In situations a fast response is required, FFwO should be used. For all other cases, QoS-FFwO is the best choice among the six WA algorithms considered. All QoS-aware WAs can be applied to other routing algorithms to improve the search for a better lightpath with higher probability of fulfilling the QoS requirements. This is a subject of onging work. ACKNOWLEDGEMENT

−1

10

J. He and M. Brandt-Pearce were supported by the National Science Foundation under grant CNS-0520060. S. Subramaniam was supported by NSF Grant CNS-0519911.

−2

10

R EFERENCES −3

10

FF RP FFwO QoS−RP QoS−FF QoS−FFwO

−4

10

−5

10

−55

−50

−45

−40

−35

−30

X , dB sw

Fig. 9. Call blocking probability caused by unsatisfactory BER for six WA algorithms with BER and latency guarantees.

FFwO has the nearly the same delay as QoS-FF as shown in Fig. 8. In Fig. 8, the BPs of timeout for RP, FF, and FFwO are always negligible. FFwO has the least BER blocking, followed by RP and FF in low or medium crosstalk situations. From Fig. 9, we can see that wavelength ordering does not provide any benefits when Xsw is large so that FFwO converges to FF and QoS-FFwO converges to QoS-FF. QoS-RP and RP still benefit from their randomness property in all situations. But QoS-RP is the worst QoS-aware algorithm, suffering from high BER blocking as shown in Fig. 9. The results here are consistent with previous results. QoSaware WA algorithms are superior to QoS-guaranteed WA algorithms. QoS-FFwO is the recommended algorithm for normal levels of crosstalk power.

[1] B. Mukherjee, “WDM optical communication networks: progress and challenges,” IEEE Journal on Selected Areas in Communications, vol. 18, no. 10, pp. 1810–1824, October 2000. [2] B. Ramamurthy, D. Datta, H. Feng, J. Heritage, and B. Mukherjee, “Impact of transmission impairments on the teletraffic performance of wavelength-routed optical networks,” J. Lightwave Technol., vol. 17, no. 10, pp. 1713–1723, Oct. 1999. [3] J. He and M. Brandt-Pearce, “RWA using wavelength ordering for crosstalk limited networks,” in Proceedings of the IEEE/OSA Optical Fiber Conference (OFC), Anaheim, CA, USA, Mar 2006. [4] ——, “Dynamic wavelength assignment using wavelength spectrum separation for crosstalk limited networks,” in Proceedings of the IEEE International Conference on Broadband Networks (Broadnets), San Jose, CA, USA, 2006. [5] H. Zang, J. Jue, and B. Mukherjee, “A review of routing and wavelength assignment approaches for wavelength-routed optical WDM networks,” Optical Networks Magazine, no. 1, pp. 47–60, 2000. [6] T. Deng, S. Subramaniam, and J. Xu, “Crosstalk-aware wavelength assignment in dynamic wavelength-routed optical networks,” in Proceedings of the IEEE International Conference on Broadband Networks (Broadnets), Oct. 25–29, 2004. [7] Y. Pointurier, M. Brandt-Pearce, T. Deng, and S. Subramaniam, “Fair QoS-aware adaptive Routing and Wavelength Assignment in all-optical networks,” in Proceedings of the IEEE International Conference on Communications (ICC), Istanbul, Turkey, June 2006. [8] I. Fonseca, J. R. Almeida, H. Waldman, and M. Ribeiro, “Meeting optical QoS requirements with reduced complexity in dynamic wavelength assignment,” in Proceedings of the IEEE International Conference on Broadband Networks (Broadnets), Oct. 2004, pp. 184–193. [9] R. Ramaswami, “Optical networking technologies: What worked and what didn’t,” IEEE Communications Magazine, vol. 44, no. 9, pp. 132– 139, Sept. 2006. [10] G. Agrawal, Fiber-Optic Communication Systems, 3rd ed. Wiley, New York, NY, USA., 2002.

Suggest Documents