Chapter 3 Utility-based Fuzzy Wavelength Assignment (UFWA)
3.3 Wavelength Selector (WS)
The WS will choose the wavelength with the largest degree among the Ai, for i = 1, 2, …, M. And then the WS will inform the WC and OSM which wavelength is assigned to the DB, and schedule the corresponding FDL, denoted by Fx, to the DB.
21
Chapter 4 Simulation Results and Discussion
4.1 Simulation Environment
The link capacity of each fiber is 10 Gbps, and each fiber has 8 wavelengths (W=8). Therefore, each wavelength has the transmission rate equivalent to 1.25 Gbps.
In order to support QoS, there are two kinds of traffic types, including the real time traffic and the non real time traffic. The real time traffic has its own delay sensitive nature, so it cannot wait for a long time to aggregate a large burst. Thus, the real time traffic would have shorter average burst length than non real time traffic. In the simulation, the average burst length of the real time traffic is 8 μs (equivalent to 10Kb), and the average burst length of the non real time traffic is 32 μs (equivalent to 40Kb). And both the bursts inter-arrival time and the burst lengths follow the Pareto distribution with the parameter k=1.5 [17]. In addition, each incoming burst would have the probability 0.5 to be the real time traffic and the probability o.5 to be the non real time traffic. The number of the remaining hops will be generated uniformly between 1 and 10. And there are 10 different kinds of length of FDLs (K=10), where the basic delay time is 10 μs (D=10).
22
4.2 Compared Algorithms
There are two other conventional wavelength assignments used in PPJET [3, 7, 8, 9], called preemptive latest available unused channel with void filling (PLAUCVF) and efficient preemption-based channel scheduling (EPCS). For the case FDLs are not used and preemption does not occur, these two algorithms will first check the duration of [Ta, Ta+B] to find whether the duration is available or not. If the duration, [Ta, Ta+B], is available on several wavelengths, the PLAUC and the EPCS will select the wavelength with the smallest void. If there are two or more wavelengths with the smallest void, the two algorithms will choose one randomly. Once [Ta, Ta+B] is not available on any wavelengths, FDLs would be considered to be used to delay the associated DB. In the case of using FDLs, the two algorithms will find the suitable unused shortest FDL. While there are not any suitable FDLs could be used and the associated DB could preempt some assigned DB, whose reserved durations overlap [Ta, Ta+B], on some wavelengths, the PLAUCVF and the EPCS would choose different wavelengths to assign the associated DB based on their own algorithms. The PLAUCVF would select the wavelength with the smallest void to assign, while the EPCS will choose the wavelength with the smallest length of the preempted bursts from the preemptive region. The PLAUCVF divides all W wavelengths into two parts, called the shard region (contains W-m wavelengths) and the preemptive region (contains m wavelengths). The wavelengths in the shared region will view all kinds of traffic as the same type (which means preemption will not occur in the shared region), while the wavelengths in the preemptive region can still adopt preemption. The following figure 4.1 shows the cases of not using FDLs and using FDLs while preemption does not occur and the figure 4.2 and 4.3 show the cases of preemption.
23
Fig. 4.1 (a) The case of not using any FDLs (b) The case of using a suitable FDL
In the figure 4.1 (a), the associated DB could be assigned on wavelength #1, #3, and #4 without using any FDLs. And the two algorithms would choose the wavelength #1 because it has the smallest void among the others. In the figure 4.1 (b), the associated DB cannot find any available wavelengths at the duration [Ta, Ta+B], so the FDL would be considered. While a suitable unused shortest FDL would be used, the PLAUCVF and the EPCS will choose the wavelength #3 because it has the smallest void among the others.
Fig. 4.2 The case of preemption of the PLAUCVF
When there are not any suitable FDLs to be used and the associated DB has the higher priority, the preemption could be adopted. In the figure 4.2, the PLAUCVF could choose either the wavelength #3 or the wavelength #4 because the preemption
24
cannot be adopted on the other two wavelengths due to the same priority. The PLAUCVF would finally select the wavelength #4 due to the smallest void.
Fig. 4.3 The case of preemption of the EPCS
In the figure 4.3, the EPCS has the preemptive region with the wavelength #1, #2 and #3 and the shared region with the wavelength #4. The EPCS could choose either the wavelength #2 or the wavelength #3 because the preemption cannot be adopted on the other two wavelengths due to the same priority and the shared region scheme. The MPLAUC would eventually select the wavelength #2 due to the smallest length of the preempted bursts.
25
4.3 Simulation Results
In this section, we will use some figures to show the better performance of the utility-based fuzzy wavelength assignment (UFWA) among the PLAUCVF and the EPCS.
4.3.1 Throughput
Fig. 4.4 illustrates the system throughput versus the traffic load intensity. The traffic load intensity is the normalized traffic by the capacity of each fiber, 10Gbps. It can be found that the throughput is increasing more slowly. It is because when the traffic load intensity is getting larger, the duration of bandwidth requests would overlap more frequently, and FDLs could be not available to delay any DBs anymore.
Therefore, there are lots of bursts could be blocked or preempted, and that will lead to much higher burst loss probability (BLP) and lower throughput which is compared to the offered load. Besides, the proposed UFWA would provide higher throughput (when M=4, M=8) than the two compared algorithms, especially at high traffic load intensity (when traffic load intensity equal to 0.8, 0.9 and 1). For example, higher than MPLAUC-VF about 300Mbps and higher than EPCS about 150Mbps at traffic load intensity equal to 1. It is because the UFWA considering about four important parameters, preemption index (Rw), the length of the used FDL (Fw), void (Gw), and utilization (Uw). Rw avoids the occurrences of preemption and longer preemption. Fw , Gw ,and Uw help make wavelengths arranged more tightly. Other than the proposed UFWA, the PLAUCVF and the EPCS choose the shortest FDL to delay the DB while UFWA would make a balance between delays and voids to make wavelength arranged more tightly. Besides, the PLAUCVF does not consider the length of the preempted
26
bursts as a factor, while the EPCS does, and that lead the former one has worse performance than the latter one in throughput. The EPCS uses not only the length of the preempted bursts as the factor but also shared region to protect the longer but lower prioritized DB (non real time traffic in this paper) to make better throughput than the PLAUCVF. As for the utility method, which is used only the utility function to make wavelength assignments, it would avoid choosing the longer FDL to delay the real time traffic due to the delay sensitive nature. Therefore, utility method and even UFWA (M=2) would rather select the preemption operation not using longer FDL. That would cause too much preemption to have better throughput.
5.9 6.4 6.9 7.4 7.9 8.4 8.9
0.6 0.7 0.8 0.9 1
Throughput (Gbps)
Traffic load intensity PLAUCVF
EPCS Utility UFWA(M=2) UFWA(M=4) UFWA(M=8)
Fig. 4.4 Throughput (Gbps)
27
4.3.2 Total length of the preempted bursts
Fig. 4.5 shows the total length of the preempted bursts versus the traffic load intensity. It can be found that the total length of the preempted bursts is increasing when the traffic load intensity is getting larger. In addition, the utility method and the UFWA (M=2) indeed preempt too much bursts to have a better throughput discussed in the section 4.3.1. The PLAUCVF also preempts longer length because it does not take the preempted length into consideration while the EPCS preempts shorter length.
0 200 400 600 800 1000 1200
0.6 0.7 0.8 0.9 1
Total length of the preempted bursts (Mb)
Traffic load intensity PLAUCVF
EPCS Utility UFWA(M=2) UFWA(M=4) UFWA(M=8)
Fig. 4.5 Total length of the preempted bursts
28
4.3.3 Total length of the blocked bursts
Fig. 4.6 shows the total length of the blocked bursts versus the traffic load intensity. It can be found that the total length of the blocked bursts is increasing when the traffic load intensity is getting larger. In addition, the utility method and the UFWA (M=2) block the two shortest bursts length. It is because these two methods preempt too much DBs and that would let wavelengths be more “empty” than the others. Thus, the more empty wavelengths could accept more bandwidth requests. By the way, the UFWA (M=4, M=8) blocks less length that the EPCS while the EPCS preempts small total length of the DBs as the UFWA (M=4), and slightly larger than the UFWA (M=8). It is because the UFWA considers preempted length, void, and utilization, and these parameters could make wavelengths arranged more tightly to let future DBs have higher opportunity to make bandwidth reservation.
0
Total length of the blocked bursts (Mb)
Traffic load intensity
Fig. 4.6 Total length of the blocked bursts
29
4.3.4 The number of the preempted bursts
Fig. 4.7 shows the number of the preempted bursts versus the traffic load intensity. It can be found that the number of the preempted bursts is increasing when the traffic load intensity is getting larger. In addition, the PLAUCVF and the EPCS have smaller number of the preempted bursts. It is because these two wavelength assignments choose the suitable shortest unused FDL to delay DB, and that would cause larger void while the proposed UFWA could select the slightly longer FDL but make smaller void. In other words, the UFWA makes wavelength arranged more tightly. The UFWA’s characteristic lets DBs have higher chance to make bandwidth reservations but larger number of preempted bursts. Once preemption occurs, lots of number DBs could be preempted. But we should notice that the proposed still have higher throughput than the PLUCVF and the EPCS.
0
Fig. 4.7 The number of the preempted bursts
30
4.3.5 The number of the blocked bursts
Fig. 4.7 shows the number of the blocked bursts versus the traffic load intensity.
It can be found that the results match the reason in the section 4.3.2. The longer length of the preempted bursts, the higher chance of accepting bandwidth requests. Therefore, the utility method and the UFWA (M=2) will have the two smallest number of the blocked bursts.
0 1000 2000 3000 4000 5000 6000
0.6 0.7 0.8 0.9 1
The number of the blocked bursts
Traffic load intensity PLAUCVF
EPCS Utility UFWA(M=2) UFWA(M=4) UFWA(M=8)
Fig. 4.7 The number of the blocked bursts
31
4.3.6 Burst loss probability (BLP)
Fig. 4.8 shows the burst loss probability versus the traffic load intensity. The BLP is calculated by the number of the non-preempted bursts plus the number of the non-blocked bursts over the total number of the bursts. It can be found that the proposed UFWA has higher BLP than the other two wavelength assignments. It is because the UFWA considers preempted lengths and preempts shorter length, so that wavelengths do not have too much available duration to accept bandwidth request. In addition, the UFWA attends to preempt larger number due to tight arrangement by considering the other three parameters. But we should notice that even the higher BLP, the UFWA still have higher throughput.
0
32
4.3.7 BLP of the real time traffic
Fig. 4.9 shows the BLP of the real time traffic versus the traffic load intensity.
Compared to the total BLP, it can be proved that these three wavelength assignments can support QoS, where the real time traffic has much lower BLP than the non real tine traffic. Although they all have low BLP for the real time traffic, the EPCS has relatively higher BLP than the others. The reason of having higher BLP is that the EPCS uses the shared region to protect the non real time traffic not to be preempted.
Obviously, the shared region can help improve the throughput due to longer non real time bursts but it does lead to higher BLP of the real time traffic.
0 0.00005 0.0001 0.00015 0.0002 0.00025 0.0003 0.00035
0.6 0.7 0.8 0.9 1
BLP
Traffic load intensity PLAUCVF
EPCS Utility UFWA(M=2) UFWA(M=4) UFWA(M=8)
Fig. 4.9 BLP of the real time traffic
33
4.3.8 Average delay of the real time traffic
Fig. 4.10 illustrates the average delay of the real time traffic versus the traffic load intensity. It can be found the delay is getting larger while the traffic load intensity is getting stronger. It is because the more bandwidth requests, the requesting duration will have more chance to be overlapped, so that we would use FDLs to delay the DBs to find some available duration more frequently. The more frequently we use FDLs, the shorter ones could be not available any more, longer FDLs then would be used.
Besides, the UFWA make a balance between the used lengths of FDLs and the voids, so that the UFWA would use longer FDLs to have smaller voids to make wavelengths arranged more tightly. That is why the UFWA would use longer delay than the others.
Although the UFWA uses longer delay, we should realize that the usage of FDLs is based on a basic delay time D. So even the longer FDL the UFWA uses, it is just a basic delay time larger than the others.
0
Fig. 4.10 Average delay of the real time traffic
34
4.3.9 Limited Preemption UFWA
To observe the simulation result of the BLP of the real time traffic, we can find that BLP of the real time traffic is too low. Considering the nature of the shorter burst length of the real time traffic, we proposed a limited preemption scheme to modify the proposed UFWA to try to improve the throughput, called limited preemption UFWA (LP-UFWA). While doing wavelength pre-assignment function on the wth wavelength and the preemption index (Rw) exceeds a threshold, Pth, we would set Rw equal to -1 not the original value, where -1 means the associated DB is blocked on the wth wavelength. The following two figures, fig.4.11 and fig. 4.12 will show the improvement of the throughput and the associated BLP of the real time traffic. And the simulation results will be based on a threshold, Pth, which is equal to 1.5, equivalent to 15 time longer preempted length than the real time traffic. It can be found that the limited preemption does improve the throughput about 150 Mbps at traffic load intensity equal to 1. Especially for the utility method and the UFWA (M=2), the limited preemption scheme limit the occurrences of preemption, and that would overcome the weakness of preempting too much characteristic of the two algorithms. Fig. 4.12 shows the BLP of the real time traffic versus the traffic load intensity. Even the BLP is increasing dramatically compared to the original proposed UFWA, it is still only about 0.2% when the traffic load intensity equal to 1. Therefore, the LP-UFWA can maintain BLP of the real time traffic at a low level but improve the throughput.
35
Fig. 4.11 Throughput of the Limited Preemption UFWA (LP-UFWA)
0.00%
Fig. 4.12 The BLP of the real time traffic of the LP-UFWA
36
Chapter 5 Conclusions
The paper proposed the utility-based fuzzy wavelength assignment (UFWA) to not only support QoS, but also achieves higher throughput. The UFWA uses the suitable wavelength concentrator (SWC) to reduce the number of the placements of the fuzzy wavelength evaluator (FWE). The SWC also does the wavelength pre-assignment function on all W wavelengths, so that we can let all wavelengths to find their own scheduled ways other than the PLAUCVF and the EPCS. In addition, the UFWA uses four parameters, including preemption index, the used length of the FDL, void, and utilization to describe the wavelength pre-assignment results on all wavelengths. By considering these four parameters, the UFWA can use the utility function and the fuzzy logic system to make a balance between the used lengths of FDLs and the voids, and preempt bursts as short as possible. The fuzzy logic system can help us to classify these four parameters into several linguistic variables using associated fuzzy membership functions. Then we can utilize fuzzy rules to evaluate each wavelength’s adequate degree. To observe the simulation results, we find when M=4, the UFWA achieves almost the same high throughput as M=8 but we only have
to place half of the number of the FWEs than the UFWA when M=8. Besides, M=4 has shorter delay due to the more serious protection of using FDLs by the utility function.
37
Moreover, due to the relatively low BLP of the real time traffic and the nature of delay sensitive of the real time traffic (shorter burst length), we proposed another algorithm, called the limited preemption UFWA (LP-UFWA) to achieve the higher throughput but still keep the low BLP of the real time traffic about 0.2% even when traffic load intensity equal to 1.
38
Bibliography
[1] J. White, “Optical Burst Switching for Next Generation Transport Networks”, Mar. 2002
[2] C. Qiao, “Labeled Optical Burst Switching for IP-over-WDM Integration”, IEEE COMMUN MAG, Vol. 38, no. 9, pp. 104-114, Sep. 2000
[3] Y. Chang, C. Qiao, and X. Yu, “Optical Burst Switching: A New Area in Optical Networking Research”, Network IEEE, Vol. 18, no. 3, pp. 16-23, May 2004
[4] K. J. Barker, “On the Feasibility of Optical Circuit Switching for High Performance Computing Systems”, Supercomputing, 2005, Nov. 2005
[5] T. S. El-Bawab and S. Jong-Dug, “Optical Packet Switching in Core Networks:
between Vision and Reality”, Communication Magazine, IEEE, Vol. 40, no. 9, pp.60-65, Sep. 2002
[6] C. Qiao, and M. Yoo, “Optical burst switching (OBS) – A New Paradigm for An Optical Internet”, Journal of High Speed Networks, Vol. 8, no. 1, pp. 69-84, 1999
[7] J. Teng, and G. N. Rouskas, “A Detailed Analysis and Performance Comparison of Wavelength Reservation Schemes for Optical Burst Switched Networks”, Photonic Network Communications, Vol. 9, no. 3, pp. 311-335, May 2005
39
[8] K. Dolzer, C. Gauger, J. Sath and S. Bodamer, “Evaluation of Reservation Mechanisms for Optical Burst Switching”, International Journal of Electronics and Communications, Vol. 55, no. 1, pp. 18-26, 2001
[9] A. Kaheel and H. Alnuweiri, “A Strict Priority Scheme for Quality-of-Service Provisioning in Optical Burst Switching Networks”, ISCC, Vol. 1, pp. 16-21, Jun. 2003
[10] J. S. Turner, “Terabit Burst Switching”, Journal of High Speed Networks, vol. 8, no. 1, pp. 3-16, Jan. 1999
[11] Y. Xiong, M. Vandenhoute, and H. C. Cankaya, ”Control Architecture in Optical Burst-Switched WDM Networks”, Selected Areas in Communications, Vol.
18, no. 10, pp. 1838- 1851, Oct 2000
[12] J. Li, C. Qiao, and Y. Chang, ”Recent Progress in the Scheduling Algorithms in Optical-Burst-Switched Networks”, Journal of Optical Networking, Vol. 3, no.
4, pp. 229- 241, Mar. 2004
[13] A. Rostami, and A. Wolisz, “Modeling and Synthesis of Traffic in Optical Burst-switched networks”, Lightwave Technology, Vol. 25, no. 10, pp.
2942-2952, Oct. 2007
[14] A. Rostami and A. Wolisz, “Modeling and Fast Emulation of Burst Traffic in Optical Burst Switched Networks,” ICC, vol. 6, pp. 2776–2781, 2006
[15] T. Tachibana and S. Kasahara, “Burst-cluster Transmission: Service Differentiation Mechanism for Immediate Reservation in Optical Burst Switching Networks” , Communication Magazine, IEEE, Vol. 44, no. 5, pp.
46-55, May, 2006
[16] B. C. Kim, S. H. Lee, Y. S. Choi and Y. Z. Cho, “An efficient preemption-based channel scheduling algorithm for service differentiation in OBS networks”, Computer Communications 29 (2006) 2348–2360
40
[17] J. M. Pitts and J. A. Schormans, “Introduction to IP and ATM Design Performance: With Applications Analysis Software, Second Edition”, John Wiley & Sons, 2000
41
Vita
He-Jyun Lin was born in 1985 in Kaohsiung, Taiwan. He received the B.E. and M.E. degree in the department of communication engineering, college of electrical and computer engineering from National Chiao-Tung University, Hsinchu, Taiwan, in 2007 and 2009, respectively. His research interests include optical networks.