iii. Simulation Results and Discussions
C. Simulation Results
We show the performance of Q-DBA method. The traffic arrival rates are set as follows:
Fig. 15 illustrates the average video dropping probability versus system load in EPON. The dropping probability in Q-DBA is zero due to the video packets with the problem of delay requirement are served earlier by raising their priorities when the system load is less than 0.8, but, the dropping probability exceeds the video dropping probability requirement due to the limited capacity of fiber link when the system load is larger than 0.8. Unfortunately, it also can be found that the dropping probability in DBAM cannot be guaranteed due to the maximum window which is used to allocate the bandwidth of video packets cannot totally support the burst arrival. Besides, when
improved 100%.
Figure 15. Average video dropping probability versus system load in EPON
Fig. 16 illustrates the blocking probability of data packets versus system load in EPON. The blocking probabilities both in Q-DBA and DBAM are zero when the traffic load is less than 0.8. However, when the system load is larger than 0.8, the blocking probability in DBAM increases greatly than that in Q-DBA. It is due to that the data burst arrival cannot be served within a short time in DBAM and the allocated bandwidth is not limited and the data packets which reach the waiting bound would be sent early in Q-DBA. It is shown that when the system load is 0.8, the blocking probability in Q-DBA is improved 100% whether in case one or case two.
Fig. 17 illustrates the starvation ratio of data packets versus traffic load in EPON.
It is shown that when the system load is less than 0.7, the starvation in both Q-DBA and DBAM do not occur, but, when the system load is larger than 0.9, the starvation occurs due to that the burst arrival of data packets may not be instantly supported as both video and data packets arrival rate increases. However, the starvation ratio is still 60% improved by Q-DBA in case one, and it also 60% improved by Q-DBA in case two due to raising the priority of data packets which will reach their waiting bound in Q-DBA. Therefore, the starvation in DBAM happens earlier than that in Q-DBA.
Fig. 18 shows the average voice delay time versus system load in EPON. It can be seen that the average voice delay in DBAM increases with the increasing of system load and one in Q-DBA does not increase obviously. It is due to the Q-DBA allocates total bandwidth to all ONUs by the step of residual bandwidth allocation, and the voice packets have sufficient bandwidth to transmit the reported voice packets and new arrival voice packets. However, when system load is larger than 0.8, the average voice delay in DBAM is less than that in Q-DBA. Since the report message in DBAM includes prediction, the requirement can be satisfied as much as possible. Under this circumstances, the voice delay in Q-DBA is large than DBAM, but the voice delay requirement and the dropping probability are still satisfied.
Figure 16. Average data data blocking probability versus system load in EPON
Figure 17. Average data starvation ratio versus system load in EPON
Fig. 19 shows the average video delay time versus system load in EPON. The average video delay in Q-DBA is far from the video delay requirement when the system load is below 0.8, whereas, one is close to the video delay requirement when the system load exceeds 0.8. However, the average video delay in DBAM increases almost smoothly with the increasing of system load due to the maximum window and the prediction. Furthermore, the average video delay is in the case one (two) of Q-DBA is improved 78% (77%) due to the video packets with a higher priority than data packets. Besides, the video packet with the problem of delay requirement could be transmitted with a higher priority. Thus, the delay can be decreased and the dropping probability can be also satisfied even if the burst arrival occurs.
Figure 18 Average voice delay time versus system load in EPON
Figure 19 Average video delay time versus system load in EPON
Figure 20 Average data delay time versus system load in EPON
Fig. 20 shows the average packet delay time versus system load in EPON. The average data delay in case one is less than that in case two due to the less data arrival rate. Since the Q-DBA allocates bandwidth based on the requirement of queue occupancy and the burst arrival can be transmitted more easily, and thus the delay of
data packets can be small. Besides, due to the Q-DBA considers the condition of waiting bound, the data packets dose not violate the delay bound in Q-DBA as early as that in DBAM does. Also, Q-DBA improves the average data delay of the case one (two) by 98% (99%)when the system load is 0.6.
Figure 21 Fairness index of average data delay versus system load in EPON
Figure 22 Overall fairness index of data packets versus system load in EPON
Figure 23 System utilization versus system load in EPON
Fig. 21 illustrates the fairness index of average data delay versus system load in EPON. It can be found that the fairness index of average data delay in Q-DBA and DBAM is bear to 1. It is due to the Q-DBA consider all ONUs’ condition to allocated bandwidth. Furthermore, in DBAM, all ONUs have their maximum window to transmit packets according to the SLA. In this way, the fairness index of average data delay whether in Q-DBA or in DBAM is close to zero. It also can be found that the fairness index of average data delay in case two of DBAM varies greatly because the packet arrival rates are not the same, and the maximum window cannot suitably support the different traffic load.
Fig. 22 illustrates the overall fairness index of data packets versus system load in EPON. The fairness index of data packets in both Q-DBA and DBAM is close to 1.
due to the DBAM uses a maximum window to allocate packets according to the SLA, and the Q-DBA allocates packets based on the requirement of all ONU. The overall fairness index of data packets in Q-DBA decreases with the increasing of traffic load when the traffic load is larger than 0.7. Since the data packets in Q-DBA are allocated in two different priority, the overall fairness is little small than 1 when the traffic load is high. In DBAM, the overall fairness index in case one is independent of different system loads. Although the DBAM in case one has a maximum window to allocate packets, the burst in high priority cannot be always processed quickly to result packets’ blocking and long delay time.
Fig. 23 shows the system utilization versus system load in EPON. It can be found that the bandwidth utilization in Q-DBA is better than that in DBAM. It is because the bandwidth in Q-DBA is allocated step by step to difference class rather than set a maximum window to each class in advance. In addition, because the dropping probability of video packets is high, and the maximum window does not always meet the actual traffic condition, the bandwidth utilization in DBAM is limited. It also can be found that when system load is in 0.6, the case one of Q-DBA improves the system utilization 4.4%, and the case two improves 7.3%.