Fuzzy STQ Output Rate Calculation

The STQ output rate is an important factor to determine the fairRate. It limits the amount of the transit traffic in the STQ which can be passed to the downstream node in the unit time. In the original fairness control mechanism, the STQ output rate is limited by the fairRate, too. However, if the node has not much local ingress traffic, the STQ output rate should be given a high rate to accelerate the decreasing of the STQ occupancy.

Therefore, in the fuzzy fairness control mechanism, we define the method to estimate the appropriate STQ output rate.

There are two important factors to determine the STQ output rate. One is the difference of the received fairRate, denoted by drf(n), which shows that the variation of the received fairRate from the downstream node. The received fairRate can represent the capacity of the STQ in the downstream node. According to the received fairRate, the addRate of the local station will be limited. The total output transmission rate will be determined. The STQ output rate is influences the total output transmission rate. If the received fairRate increases, the total output transmission rate may increase and the STQ output rate may increase. On the contrary, if the received fairRate decreases, the total transmission rate will decrease and the STQ output rate will decrease. The other factor is the ingress traffic of the local station in the previous time unit, denoted by T(n-1), which has higher priority than the traffic in the STQ. Since the STQ is served lastly, the ingress traffic of the PTQ, Queue A, and Queue B should be served before the STQ. The amount of the traffic in PTQ, Queue A, and Queue B will directly influence the STQ output rate due to their priority. The term set for the difference of the received fairRate is defined as T(drf(n)) = {Positive High (PH), Positive Low (PL), Negative High (NH), Negative Low (NL)}; for the ingress traffic in the previous time unit which has higher priority than the

STQ is defined as T(T(n-1)) = {Big (B), Small (S)}; for the determined STQ output rate is defined as T(os(n)) = {Very High (VH), High (H), Low (L), Very Low (VL)}.

The membership functions for PH, PL, NL, and NH in T(drf(n)) are denoted by μPH(drf(n)), μPL(drf(n)), μNL(drf(n)), and μNH(drf(n)), and they are

μPH(drf(n))= f(drf(n); RL, 0.75RL, 0) (29)

μPL(drf(n))= f(drf(n); 0.25RL, 0.25RL, 0.25RL) (30)

μNL(drf(n))= f(drf(n); −0.25RL, 0.25RL, 0.25RL) (31)

μNH(drf(n))= f(drf(n); −RL, 0, 0.75RL) (32)

where RL is the link rate of the output link. The membership functions for B and S in T(T(n-1)) are denoted by μB(T(n-1)), and μS(T(n-1)), and they are μB(T(n-1))= g(T(n-1); 0.6RL, RL, 0.4RL, 0.4RL) (33)

μS(T(n-1))= f(T(n-1); 0, 0, 0.4RL) (34)

where RL is the link rate of the output link. The fuzzy inference algorithm is also min-max inference method. The defuzzification method we used is the center of area defuzzification method. The membership function for VL, L, H, and VH in os(n) are denoted by μVL(os(n)), μL(os(n)), μH(os(n)), μVH(os(n)), and they are defined as μVL(os(n))= f(os(n); 0.2 RU, 0, 0) (35)

μL(os(n))= f(os(n); 0.4 RU, 0, 0) (36)

μH(os(n))= f(os(n); 0.6 RU, 0, 0) (37)

μVH(os(n))= f(os(n); 0.8 RU, 0, 0) (38)

where RU is the unreserved rate of the output link.

The fuzzy rules of the fairRate calculation are shown in Table 3.3. We determine the STQ output rate based on the deference of the received fairRate. If the received fairRate increases, the corresponding STQ output rate will be set in a high level.

Oppositely if the received fairRate decreases, the STQ output rate will be set in a low level. The ingress traffic in the previous time unit which has higher priority than the STQ is a reference value to adjust the STQ output rate. The STQ output rate will be leveled down because of the large amount of the ingress traffic with high priority.

Table 3.3

The rule base of the STQ output rate calculation

Rule drf(n) T(n-1) o_s(n) Rule drf(n) T(n-1) o_s(n) 1 PH S VH 5 NL S H 2 PH B H 6 NL B L 3 PL S H 7 NH S L 4 PL B L 8 NH B VL

Chapter 4

Simulation Results and Discussions

4.1 Simulation Environment

In this section, a time-driven packet-based simulation is developed to show the performance of the proposed fuzzy fairness control mechanism (FFCM) presented in the previous chapter. We consider the RPR architecture with five nodes. The link bandwidth between the stations is considered to be 1Gbps. The propagation delay is set to 0.2ms.

The aggressive mode (AM) fairness control mechanism is implement to compare with the FFCM.

Three kinds of traffic are considered in the system: voice, video, and data. The voice traffic is transmitted as the Class A traffic with the highest priority, and is generated by an ON/OFF model defined in Chapter 2. The video traffic is transmitted as the Class B traffic. The main frames of the video will be transmitted as the Class B-CIR traffic, the others will be sent as the Class B-EIR. The data traffic will be transmitted as the Class C

traffic.

Figure 4.1 shows the simulation model. There are five nodes in the RPR ring which is indicated by node 1, node 2, node 3, node 4, and node 5. The corresponding link to transmit the data frames from node 1, node2, node3, and node 4 is called link 1, link 2, link 3, and link4. In this simulation model, we define four flows which indicate the ingress traffic to node 5. The flow 1, for instance, means the traffic from node 1 to node 5 which is transit in node 2, node 3, and node 4. The flow 2, flow 3, and flow 4 is similar to the flow 1.

Figure 4.1: The simulation model

Each flow consists of the fairness eligible traffic, i.e. the Class B-EIR traffic and the Class C traffic, with a mean traffic rate. The mean traffic rate is a ratio of the link capacity. If the rate is 0.1, for instance, the traffic load is 100Mbps. The Class B-EIR traffic and the Class C traffic will be generated by the bursty traffic model with the arrival rate of the High state is set to 0.9 and the arrival rate of the Low state is set to 0.1. We can adjust the change probability to generate different traffic load. The Class A and the Class B-EIR traffic will be generated be the ON/OFF model and the Sup-FRP Model. The total traffic load of the Class A and the Class B-EIR will vary in the range of 10Mpbs to

100Mbps.

There are two scenarios in our simulation, the balance mode and the unbalance mode. In the balance mode, all the flows have the same traffic load. As the traffic load of each flow is getting higher, the link 4 will be overloaded, and the node 4 will become congested. We observe the link utilization of the link 2 and link 3 to show that the fuzzy fairness control mechanism is able to control the congestion as the aggressive mode fairness control mechanism which is defined by the 802.17 standard. In the unbalance mode, the traffic load of the flow 2 and the flow 4 is fixed at 0.1, and we change the traffic load of the flow 1 and flow 3 from 0.15 to 0.4. As in the balance mode, the node 4 will become congested, too, but the fair rate will over throttle the transit traffic from the upstream node of node 4 because of the extreme low load of the node 4. In the unbalance mode, we will observe the link utilization and the access delay of each flow to test and verify the influence of the over throttle behavior; also we will show that the fuzzy fairness control mechanism can improve the performance.

4.2 Simulation Results

Figure 4.2 shows the link utilization of the link 2 and link 3 in the balance mode.

The ingress traffic load is set from 0.1 to 0.225, i.e. the system traffic load is set from 0.4 to 0.9. In Figure 4.2 (a), we can observe that the performance of the aggressive mode and the fuzzy fairness control mechanism are nearly the same. This result shows that when the system load is light and balance, both of the two fairness control mechanism work well because the node does not become congested frequently. But in Figure 4.2(b), when the load is getting higher, the utilization of the link 3 under aggressive mode is clearly lower than the utilization under the fuzzy fairness control mechanism. The reason of this

result is that the node 4 is treated as congested only when the STQ occupancy exceeds the pre-determined threshold in the aggressive mode, but the node information, such as the net input rate of the STQ and the STQ output rate of the local node, is considered to determine whether the node is congested or not in fuzzy fairness control mechanism.

Even if the STQ occupancy is in a really high position, the FFCM may not treat the node as congestion because of the low net input of the STQ and the high STQ output rate. In other words, the frequency of the congestion in fuzzy fairness control mechanism is less than in the aggressive mode.

In Figure 4.3, the utilization of the link 2 and link 3 is shown. The horizontal axis is the ingress traffic load of the flow 1 and flow 3 in the unbalance mode. Both of the ingress traffic loads of flow 2 and flow 4 are set at 0.1. In Figure 4.3(a) the utilization is similar as Figure 4.2(a) because of the low traffic load which transit by the link 2, but in Figure 4.3(b) we can observe that the utilization of the link 3 under the fuzzy fairness control mechanism is deferent obviously because of the over throttle problem.

When the node 4 is congested, it will send it’s addRate as fairRate to the upstream node firstly. In the unbalance mode, the addRate of the node 4 is less than 0.1 because of the low ingress traffic load, so the unreasonable fairRate is sent to the node 3. This fairRate causes that the ingress traffic of the node 3 is limited excessively. The utilization of link 3 decreases clearly because of the unnecessarily limitation. In our fuzzy fairness control mechanism, as the occupancy of the STQ in node 4 is getting higher, the node 4 will send a lower fairRate to its upstream node, too. To differ with the aggressive mode, the ingress traffic load and the STQ output rate is considered, and node 4 will determine a higher fairRate because of the lower local ingress traffic load and the higher STQ output rate. This behavior can estimate the appropriate fairRate more correctly and prevent the

The link utilization of link 2

the ingress traffic load of each flow

average utilization the ingress traffic load of each flow

average utilization

AM FFCM

(b)

Figure 4.2: The utilization performance in the balance mode (a) the link 2 (b) the link 3

over-throttle problem to decrease the utilization.

For the reason we described in the previous paragraph, the ingress traffic of node 3 will be limited excessively. In Figure 4.4(a), we can observe that by the access delay performance of each flow, and we notice that the access delay of the flow 1 is distinct unreasonable higher in the aggressive mode. When the congestion happens, the fairRate which is sent to all nodes in the congestion domain is the same. So the flow 1 is over throttled as the flow 3 because of the high ingress traffic load which is the same as the flow 3.

In Figure 4.4(b), we can find that the access delay of each flow is almost the same.

The fuzzy fairness control mechanism does not only prevent the over-throttle problem but also allocates the link capacity for all the ingress traffic flow more efficiently. The node with heavy traffic load will be allocated more link bandwidth to transmit the ingress traffic before the delay bound. As the result shows in Figure 4.4(b), the access delay is fairly distributed between all flows. In the aggressive mode, the access delay of the flow 1 and flow 3 are higher than the flow 2 and flow 4 because of the fairness access probability of each flow which is controlled by the fairness control mechanism. This problem is improved by FFCM because the link bandwidth is shared according to the traffic load of the flows.

The link utilization of link 2

the ingress traffic load of the flow 1 and flow 3

average utilization the ingress traffic load of the flow 1 and flow 3

average utilization

AM Fuzzy

(b)

Figure 4.3: The utilization performance in the unbalance mode (a) the link 2 (b) the link 3

average access delay of AM

the ingress traffic load of flow 1 and flow 3

access delay(ms)

the ingress traffic load of flow 1 and flow 3

access delay(ms)

Figure 4.4: The access delay of each flow in the unbalance mode (a) aggressive mode (b) the fuzzy fairness control mechanism.

Chapter 5 Conclusion

In this thesis, we propose a fuzzy fairness control mechanism to solve the problems which exist in the aggressive mode fairness control mechanism. The goal is to increase the link utilization and to share the link capacity to each node effectively. We study the architecture of node and network architecture in RPR which is defined in standard 802.17.

The proposed fuzzy fairness control mechanism is divided as three parts. First, we decide the congestion degree of the node by observing the STQ input rate and STQ occupancy.

Second, we calculate the STQ output rate by gathering the fairRate from the downstream node and the ingress traffic load information. Finally, we determine the fairRate to the upstream node to limit the traffic to the node.

The proposed fuzzy fairness control mechanism is compared to the aggressive mode fairness control mechanism. To differ with the aggressive mode, the congestion does not be determined to happen only when the STQ occupancy exceeds the pre-defined threshold, and we calculate the fairRate with more information of the node instead of setting the addRate as the fairRate firstly. As the result of that, the STQ can be utilized

effectively, and the fairRate is estimated more reasonably.

Simulation results show that the utilization of the link in the mechanism we proposed is higher than in the aggressive mode because the over-throttle problem is prevented and the congestion does not happen excessively persistently. Also the access delay of each node will become almost the same even if the ingress traffic load of each node is different. The fairness to access the RPR ring of all the nodes is improved. The fuzzy fairness control mechanism is more feasible and robust for the RPR network then aggressive mode fairness control mechanism.

Bibliography

[1] T. H. Wu and W. Way, “A novel passive protected SONET bi-directional self-healing ring architecture,” MILCOM’91, Conference Record, 'Military Communications in a Changing World', IEEE, vol.3, pp.894-900, Nov. 1991.

[2] IEEE Standard 802.17, Resilient Packet Ring (RPR) access method and physical layer specification, Sept. 2004

[3] F. Davik, M. Yilmaz, S. Gjessing, and N. Uzun, “IEEE 802.17 resilient packet ring tutorial,” IEEE Communication Magazine, vol.42, no.3, pp.112-118, March 2004.

[4] F. Davik and S. Gjessing, “The Stability of the resilient packet ring aggressive fairness algorithm,” in Proceedings of the 13^th IEEE Workshop on Local and Metropolitan Area Network (LANMAN2004), April, 2004, pp. 17-22.

[5] F. Davik, A. Kvalbein, and S. Gjessing, “An analytical bound for convergence of the resilient packet ring aggressive mode fairness algorithm,” in Proceedings of the IEEE ICC May, 2005.

[6] F. Davik, A. Kvalbein, S. Gjessing, and N. Uzun, “Improvement of resilient packet ring fairness,”

in Proceedings of the Global Telecommunications Conference, 2005, Volume 1, Nov, 2005.

[7] Y. F. Robichaud and C. C. Huang, “Improvement fairness algorithm to prevent tail node induced oscillations in RPR,” in Proceedings of the IEEE ICC May, 2005.

[8] T.J.Kim, “An enhanced fairness algorithm for the IEEE 802.17 resilient packet ring,” IEICE Transactions of Communication, May, 2005.

[9] D.H. Lee, and J.H. Lee, “A novel fairness mechanism based on the number of effective nodes for efficient bandwidth allocation in the resilient packet ring, ”IEICE Transactions of Communication, vol.E89-B, no.5, MAY, 2006.

[10] W.E. Leland, M.S. Taqqu, W. Willinger, and D. V. Wilson, “On the self-similar nature of ethernet traffic (extended version), ”IEEE/ACM Transactions on networking, vol.2, no.1, Feb, 1994.