Delay analysis of dynamic bandwidth allocation for triple-play-services in EPON

Today, triple play services may be regarded as required services in telecommunication sector. In the access network, Ethernet Passive Optical Network (EPON) is considered as one of the best solutions to provision triple-play-services. In EPON, the Dynamic Bandwidth Allocation (DBA) algorithm plays a key role to meet the Quality-of-Service (QoS) requirements. Therefore, in this paper, we conduct the delay analysis of DBA so that it can provide a better understanding for the network operators when designing and implementing a DBA algorithm. We use machine learning algorithm such as Random Forest, Decision Tree, kNN, and Naives Bayes and R-programming to build a prediction model and find the important factors affecting the packet delay in EPON system. We run extensive simulation scenarios to build appropriate dataset for training set and testing set. We found that the cycle time selection has a significant impact on the EF delay, AF delay and BE grant. We also found that there was a strong relationship between the EF delay and AF GRANT. Lastly, we conduct performance evaluation to justify what we found. Simulation results show that cycle time is indeed has significant effect on system performances.