Simulation Results and Discussions

In the simulations, settings for the environment include 10 Gbps link capacity, 100 μs propagation delay between stations, 4 Mbytes STQ size, and 100 μs agingInterval.

The value of the stqHighthreshold is 1 Mbytes and the value of the stqLowthresh-old is 0.5 Mbytes. Simulations for the proposed FLAG, DBA with moving average technique (DMA), DBA [24, 32, 33], and AM [17, 22] also conducted for perfor-mance comparison. Simulation results are recorded per agingInterval. Also, assume that the reserved bandwidth is zero, and only fairness eligible (FE) traffic flow is considered.

Fig. 3.9(a) shows a small parking lot scenario where there are 5 (0∼ 4) greedy

stations, and Figs. 3.9(b), 3.9(c), 3.9(d) and 3.9(e) present the throughput of each flow by AM, DBA, DMA, and FLAG, respectively. This small parking lot scenario assumes that flows are generated from station 0, 1, 2, and 3 but terminated at station 4. The propagation delay is small. It can be seen that FE flows of AM, DBA, DMA, and FLAG take 49ms, 14ms, 13.5ms, and 7ms to stabilize, respectively. Thus FLAG improves by 7 times over AM and by 2 times over DBA, in the convergence time of traffic flows. The reasons are given as follows. The fuzzy logics provides a robust mathematical method to solve problems which are complicated to find a proper mathematical model for them. Especially, the FLAG contains sophisticated functional blocks, which combine advantages of AM and DBA. It fine-tunes the so-called p-fairRate generated by AFC, according to the congestion degree softly determined by the FCD using the fuzzy logic and the effective fuzzy rules designed in FFG by expert’s domain knowledge. On the other hand, the DBA and DMA generate the local fairRate depending only on the short-term (average) arriving FE traffic flow, or equivalently the change rate of the STQ, without considering the STQ occupancy which usually used to determine the congestion degree of station given in [17]. This would incorrectly limit the amount of the passing transit FE traffic flow to the next station and cause DBA make error decision. For example, if the amount of the short-term arriving transit FE traffic flow is large but the STQ occupancy of a station is short, the station should not seriously regulate the FE traffic flow of its upstream stations. Also, AM generates a local fairRate which is equal to the added FE traffic flow rate of the station to regulate the flow when the station is in congestion. AM immediately sets the advertised fairRate as FullRate to allow the upstream stations to un-limitedly send traffic flow when the congestion is

Figure 3.9: (a) Small parking lot scenario with greedy traffic, and the throughput of (b) AM, (c) DBA, (d) DBA with moving average (DMA), and (e) FLAG.

released. This too-much variation of the advertised fairRate would cause the station congestion again and thus make the flow of AM damping the longest.

Fig. 3.10(a) shows a large parking lot scenario where there are containing 8 (0 ∼ 7) greedy stations, and Figs. 3.10(b), 3.10(c), 3.10(d) and 3.10(e) present the throughput of flow(0, 7), flow(2, 7), flow(4, 7), and flow(6 ,7) at station 7 by AM, DBA, DMA, and FLAG, respectively. This scenario differs from the previous

one of Fig. 3.9 in that the propagation delay would be large. It can be seen that the FLAG and the AM take 11ms and 27ms to stabilize the flows, respectively;

unfortunately, DBA and DMA take quite a long time to stabilize the traffic flows.

It is because that DBA computes the number of the effective IA flows referring to both the short aggregating traffic (per agingInterval) and the pervious local fairRate to generate the current local fairRate. However, due to the large propagation delay, the correlation between the short aggregating traffic and the pervious local fairRate becomes low. Therefore, DBA cannot generate a correct local fairRate to regulate flows. Thus the flows oscillate and converge slowly; the convergence time takes about 0.15s which is not shown here. The DMA uses the moving average technique to lessen the effect of propagation delay. The flow oscillation of the DMA is half smaller than the DBA but still exists. Since without considering the STQ occupancy for the congestion degree of station, the DMA incorrectly limits the amount of the passing transit FE traffic flow to the next station. On the other hand, the FLAG can correctly generate the p-fairRate to meet the RIAS fairness and diminish the effect of the propagation delay to some extent. Also, the FLAG finely adjusts the p-fairRate to a precise local fairRate according to both the congestion degree and the effective fuzzy rules well designed by domain knowledge. The main reason that AM in this scenario takes less time to stabilize all flows than AM in the previous scenario shown in Fig. 3.9(b) is given below. Since, here in Fig. 3.10(a), there are more stations with greedy traffic, more aggregated traffic per agingInterval will be caused. This more aggregated traffic and the larger propagation delay would make the station congestion always occur earlier. Afterwards, the station would not have the chance to set the advertised fairRate as FullRate. Thus the convergence time is

shorter.

Figure 3.10: (a) Large parking lot scenario with greedy traffic, and the throughput of (b) AM, (c) DBA, (d) DMA, and (e) FLAG.

Fig. 3.11(a) shows a large parking lot scenario where there are containing 8 (0 ∼ 7), such as in Fig. 3.10(a) but with various finite traffic demands, greedy stations, and Figs. 3.11(b), 3.11(c), 3.11(d), and 3.11(e) present throughputs of flow(0, 7), flow(2, 7), flow(4, 7), and flow(6, 7) at station 7 by AM, DBA, DMA, and FLAG, respectively. Assume that flow(0, 7) and flow(1, 7) require 2.1 Gpbs,

Figure 3.11: (a) Large parking lot scenario with greedy traffic, and the throughput of (b) AM, (c) DBA, (d) DMA, and (e) FLAG in a large parking lot scenario with various finite traffic flows.

7) require 1.0 Gbps. It would be facts that station 6 will be the first one to incur congestion, and the added FE traffic flow to network at each station cannot always match its received fairRate due to the finite traffic demand at each station. Also, flow(0,7) and flow(1,7) will have the highest throughput when station 6 is in free-congestion or the remaining bandwidth is large because of their largest required traffic demands. It can be seen that at the first beginning, all flows just oscillate slightly, and then AM, DBA, and DMA oscillate all the ways, while FLAG can make all flows converge but takes 30 ms. It is because that FLAG indeed diminishes the effect of the propagation delay and generates the correct local fairRate at each agingInterval. Also, since each traffic flow is with different finite traffic demand and is much less than that of the greedy case in Fig. 3.10(e), the damping amplitude is smaller than that in Fig. 3.10(e). Moreover, the FLAG stably realizes the RIAS fairness and has higher throughput by about 2.8%, 3.5%, and 2.4% than AM, DBA and DMA, respectively. On the other hand, the advertised fairRate by AM is often set as FullRate in this scenario because the bandwidth of the total demand traffic is 10.2 Gbps, slightly higher than the link capacity but much less than that of the greedy case in Fig. 3.10(b). In this situation, the aggregated traffic per agingInterval would be smaller, and the congestion, if any, could be solved by AM most of time. Thus, the flows by AM oscillate always and the flow(0,7) seriously oscillates due to its largest traffic demand. By DBA, its generation accuracy of local fairRate is susceptible to the propagation delay, as seen in Fig. 3.10. Also, in this scenario, station 0 and station 1 are the farthest ones to station 6 and flow(0,7) and flow(1,7) are with the largest traffic demand. These facts result in that flow(0,7) and flow(1,7) cannot be regulated by the station 6 quickly. This violent varying

Figure 3.12: The throughputs of (a) AM, (b) DBA, (c) DMA, and (d) FLAG in a large parking lot scenario containing 8 stations, where each flow is with truncated Pareto traffic model.

aggregation traffic per agingInterval and the effect of the propagation delay thus result in DBA generating the local fairRate improperly. Notice that if flow(0,7) requires less traffic demand, the oscillation amplitude of flows will be smaller. The DMA has the same phenomenon but its performance is better than DBA by 1.5%

due to using the moving average technique.

Figs. 3.12 (a), 3.12 (b), 3.12 (c), and 3.12 (d) present throughputs of flow(0, 7), flow(2, 7), flow(4, 7), and flow(6, 7) at station 7 by AM, DBA, DMA, and FLAG, respectively, in a large parking lot scenario containing 8 stations as in Fig. 3.10(a),

where each flow is with truncated Pareto traffic model [52]. Assume that flow(0, 7) and flow(1, 7) require 2.1 Gbps, flow(4, 7) and flow(5, 7) require 1.5 Gbps, and flow(2, 7), flow(3, 7) and flow(6, 7) require 1.0 Gbps. We can see that the phenomena of all flows are the same as those in Fig. 3.11, where all algorithms oscillate all the ways but FLAG makes all flows be with the smallest oscillation comparing with the other three algorithms. Thus we can claim that, due to the robustness and the sophisticate of the proposed FLAG for the fairness control, the FLAG can still perform better than the other schemes in the cases of realistic traffic models.

Fig. 3.13(a) shows an available bandwidth reclaiming scenario where there are 9 stations with finite traffic demand and a spatial reuse of flow(a,2) occurs, and Figs. 3.13(b), 3.13(c), 3.13(d) and 3.13(e) present the throughput of flow(a,2) at station and flow(0,7), flow(1,7), flow(2,7), and flow(6,7) at station 7 by AM, DBA, DMA, and FLAG, respectively. In this scenario, the flow(a, 2) requires 5.9 Gpbs, and similar to Fig. 3.11, flow(0, 7) and flow(1, 7) require 2.1 Gpbs, flow(4, 7) and flow(5, 7) require 1.5 Gpbs, and flow(2, 7), flow(3, 7), and flow(6, 7) require 1.0 Gbps. It can be seen that, just as in Fig. 3.11, at the beginning, all flows of all algorithms oscillate slightly, and finally FLAG makes all flows stabilize but takes 78 ms, while AM, DBA, and DMA oscillate all the ways. The reasons that all algorithms in this scenario behave worse than in the large parking lot scenario with various finite traffic flows, given in Fig. 3.11, are as follows. Since flow(a,2) is sunk at station 2, station 1 would have more transient FE traffic flows than station 2, where station 1 has 10.1 Gbps traffic flow maximum, while station 2 has 5.2 Gbps traffic flow maximum. This phenomenon is conversed in Fig. 3.11, where station 1 has 4.2 Gbps traffic flow maximum, while station 2 has 5.2 Gbps maximum. Therefore, the

Figure 3.13: (a) Available bandwidth reclaiming scenario with finite traffic demand, and the throughput of (b) AM, (c) DBA, (d) DMA, and (e) FLAG.

station 1 in Fig. 3.13 will more frequently and heavily regulate its station 0, which has 5.9 Gbps transient traffic flow and 2.1 Gbps local traffic flow, than the station 1 in Fig. 5 will regulate its station 0, which has only 2.1 Gbps local traffic flow.

Thus it can be believed that all flows in Fig. 3.13 would oscillate worse than in Fig.

3.11 for all schemes. Moreover, according to our computation, the throughput at station 6 by FLAG is about 0.990, which is higher than AM’s 0.825, DBA’s 0.914, and DMA’s 0.933. The reasons would be the same as those given before and are not mentioned again here.

Fig. 3.14(a) shows an available bandwidth reclaiming scenario with reuse traffic flows, where there are 9 stations with finite traffic demand and two spatial reuses of flow(a, 3) and flow(0, 3). Figs. 3.14(b), 3.14(c), 3.14(d), 3.14(e) and 3.14(f) present throughputs of flow(a, 3) and flow(0, 3) at station 3 and throughputs of flow(1, 7), flow(4, 7), and flow(6, 7) at station 7 by AM, DBA, AFC, FLAG, and M-FLAG, respectively, where the M-FLAG denotes the modified FLAG with DBA to replace AFC. In this scenario, flow(a, 3) and flow(0, 3) require 3.0 Gbps, flow(1, 7) and flow(2, 7) require 2.1 Gbps, flow(3, 7) and flow(6, 7) require 1.0 Gbps, and flow(4,7) and flow(5, 7) require 2.0 Gbps. It can be seen that, similar to the phenomena illustrated in Fig. 3.11, at the beginning, all flows of all algorithms oscillate slightly, and finally FLAG and M-FLAG make all flows stabilized and take 48 ms and 46 ms, respectively, but M-FLAG has osillations much larger than FLAG during the transitional period, while AFC converges at about 1129 ms and AM and DBA oscillate all the ways. Also, all algorithms in this scenario behave worse than in the large parking lot scenario with various finite traffic flows given in Fig. 3.11.

The reasons are as follows. Since flow(a, 3) and flow(0, 3) are sunk at station 3,

Figure 3.14: (a) Available bandwidth reclaiming scenario with finite traffic demand and two reuse traffic flows, and the throughput of (b) AM, (c) DBA, (d) DMA, (e) FLAG and (f) M-FLAG.

station 2 would have more transient FE traffic flows than station 3, where station 2 has 10.2 Gbps traffic flow maximum, while station 3 has 5.2 Gbps traffic flow maximum. This phenomenon is conversed in Fig. 5, where station 2 has 5.2 Gbps traffic flow maximum, while station 3 has 6.2 Gbps maximum. Therefore, the station 2 in Fig. 3.14 will more frequently and heavily regulate its station 1, which has 6.0 Gbps transient traffic flow and 2.1 Gbps local traffic flow, than the station 2 in Fig. 3.11 will regulate its station 1, which has 2.1 Gbps transient traffic flow and 2.1 Gbps local traffic flow. Thus it can be believed that all flows in Fig. 6 would oscillate worse than those in Fig. 5 for all algorithms. Also, that the M-FLAG has oscillations larger than the FLAG during the transitional periods shows that the AFC can indeed diminish the effect of the propagation delay once occurred in DBA.

Moreover, according to our computation, the throughput at station 6 by FLAG is about 0.993, which is higher than AM’s 0.842, DBA’s 0.921, AFC’s 0.943, and M-FLAG’s 0.988.

3.6 Concluding Remarks

In this chapter, an effective fuzzy local fairRate generator (FLAG) is proposed for resilient packet ring (RPR). The FLAG is sophisticatedly composed of three function blocks: an adaptive fairRate calculator (AFC), a fuzzy congestion detector (FCD), and a fuzzy fairRate generator (FFG). The AFC pre-generates a fairRate, which meets RIAS fairness and can diminish the effect of the propagation delay. The FCD softly detects the congestion degree of station, considering STQ queue length and its change rate which is the arriving transit FE traffic flows to STQ. Subse-quently, the FFG generates a suitable local fairRate by intelligently fine-tuning the

pre-generated fairRate, using fuzzy logics, based on the congestion degree of the sta-tion. The FLAG can make traffic flows satisfy RIAS fairness criterion and converge to an ideal fairRate in an efficient way. Simulation results show that each flow by FLAG is indeed close to the designated rate with the smallest damping amplitude and the least convergence time in the parking lot scenarios and the available band-width reclaiming scenario, compared to conventional AM, DBA, and DBA fairness algorithms. These prove that the configuration of FLAG is indeed sophisticated, where AFC pre-generates the local fairRate using the moving average technique;

FCD determines the congestion degree of station using fuzzy logics, considering not only the STQ length but also change rate of STQ length; and finally the FFG adopts the fuzzy logics and the expert’s domain knowledge to precisely generate the local fairRate by fine-tuning the pre-generated local fairRate by AFC according to the congestion degree by FCD. Also, the performance superiority of DMA over DBA proves that the moving average technique is indeed effective to diminish the effect of propagation delay on the stability of traffic flows.

Chapter 4

Intelligent Inter-Ring Route Control in Bridged Resilient Packet Rings

4.1 Introduction

The resilient packet ring (RPR) is a dual-ring-based network protocol and has been recently approved as the IEEE 802.17 Standard [17]. A RPR network consists of a clockwise (CW) and a counter-clockwise (CCW) ringlets, giving each station on the ring a full duplex connection to its neighbors. It can be used for implementing local area networks (LAN) and metropolitan area networks (MAN) at rates scalable to many gigabits per second. More than one RPR can be interconnected by a bridge which forwards packets from one RPR to another. A spatially aware sublayer (SAS), which is a part of the MAC layer, in the bridge is used to decide which ringlet interface the packet should be routed to [17, 26]. Current research on SAS, including the IEEE 802.17b Working Group, is mainly focusing on how to modify this sublayer in order to avoid flooding the entire bridged network when transmitting inter-ring packets [17, 26, 27, 28].

Settawong and Tanterdtid proposed an enhancement by using a topology

dis-covery and spanning tree algorithm [27]. The algorithm can manage traffic between rings more efficiently and can remove the need for flooding. The shortest path route controller (SPRC) was widely considered for metro rings [38, 39, 40] as it can maxi-mize the spatial reuse and thus the achievable packet throughput for uniform traffic.

However, as traffic load increases, incoming call requests could pile up at a node be-fore being processed, and these would result in a potential bottleneck in network performance [40]. Also, Heiden et. al. analyzed the capacity of bidirectional optical packet ring networks, such as RPR, which employs the SPRC for multicast hotspot traffic [41]. They found that when the multicast traffic originating at the hotspot exceeds a critical threshold, the SPRC leads to a significant capacity reduction.

Intuitively, the route selection would be closely related with the congestion degree of the ringlet so as to follow the load balancing principle. Generally, RPR uses a queue length threshold to detect the congestion and a node’s adding rate limitation to avoid the network congestion [17]. Therefore, an intuitive queue-length threshold route controller (QTRC) would be better than the SPRC. However, the correlation function between the congestion degree and these variables is nonlinear and complicated.

Recently, intelligent techniques such as fuzzy logics and neural networks have been widely applied to control nonlinear, time-varying, and well-defined systems for that fuzzy logic and neural network control can provide effective solutions with small computational complexity. Fuzzy set theory appears to be able to support a robust mathematical framework for dealing with real-world imprecision, and ex-hibits a soft behavior, which means a greater ability to adapt itself to dynamic, imprecise, and bursty environments [53]. Fuzzy and neural fuzzy implementations

of the two-threshold congestion control method and the equivalent capacity admis-sion control method were once studied in the literature [53, 54]. Results have shown that the proposed fuzzy logic and neural network approaches significantly improve system performance, compared to conventional approaches. Moreover, fuzzy logic and neural network systems are easily implemented in a chip. This will greatly reduce the computational time and make fuzzy logic and neural network control feasible for real applications.

Therefore, we propose intelligent inter-ring route control for bridged resilient packet rings in this paper. Either CW or CCW ringlet at bridge will be properly chosen for an incoming new call request from one RPR to the other. The selection is based on the load balancing principle which is in the sense that the selected ringlet would be with lower congestion degree and higher service rate [46]. An intelligent inter-ring route controller (IIRC) is designed to contain a fuzzy bridge-node congestion indicator (FBCI) to intelligently detect the congestion degree of bridge, and a pipeline recurrent neural networks (PRNN) downstream-node fairness predictor (PDFP) to effectively predict the mean received fairRate. Besides, the IIRC consists of a fuzzy router controller (FRC) to determine preference values of route of CW and CCW ringlets according to the congestion indication provided by FBCI, the predicted mean received fairRate provided by PDFP, the number of hops to destination, and the service rate of the bridge. A ringlet with a larger route preference value would be more proper to be selected. Simulation results show that the IIRC can effectively attain the load balancing property and improve the packet dropping probability (average packet delay, throughput) by 10% and 220% (13%

and 18%, 6% and 19%) over QTRC and SPRC [38], respectively, in a scenario.

This is due to the fact that the IIRC sophisticatedly detects the system congestion degree and correctly predicts the mean received fairRate using fuzzy system and neural network. Also, IIRC achieves higher throughput by 7% and 6.7% than IIRC itself but without considering the prediction of the received fairRate and without considering the amount of the reserved bandwidth as well as the equivalent capacity

This is due to the fact that the IIRC sophisticatedly detects the system congestion degree and correctly predicts the mean received fairRate using fuzzy system and neural network. Also, IIRC achieves higher throughput by 7% and 6.7% than IIRC itself but without considering the prediction of the received fairRate and without considering the amount of the reserved bandwidth as well as the equivalent capacity