Resource Management with AQM

Many algorithms have been proposed to efficiently control the congestion in the network. In this dissertation, the author first analyzes the most recent and well known AQM algorithms and evaluates their performance under heavy hybrid traffic load flows using NS-3 network simulator. For the comparative analysis, various metrics are taken into account. The author showed that AQM alone is not enough to maintain the network resources due to the traffic and network topology changes. One AQM can overcome the others in one environment and fails behind the others in another environment. According to the study, the author believed that AQM needs further assistance by the end nodes or a controller extra tool to provide the required network performance and save the network resources from the cost bufferbloat problem. Explicit Congestion Notification used with two extra parameters as extra tools to enhance one of the well-known AQM which is called Controlled Delay or CoDel. The modified CoDel was called ECN-Enhanced CoDel and tested in real network environment using Linux with TCP Cubic. ECN-Enhanced CoDel exploits the benefits of ECN marking to notify TCP sources faster than the original CoDel about the congestion. The main benefits of the proposed method are: 1) It does not affect the good properties of CoDel. 2) It significantly reduces the number of packets re-transmissions caused by the queue’s drops. 3) It reduces the observed sojourn times. 4)The implementation is incremental and does not require the deep modification of the original CoDel. To achieve the generalization of ECN with CoDel he retested it without the two parameters and with different TCP variants using P4 language. Results show that ECN and CoDel do not work well with all current TCP variants. Even ECN with CoDel could not achieve the generalization, the researcher listed as a conclusion in this dissertation the characteristics of the generalized AQM. Generalized or universe AQM should be parameters-less as much as possible; has low queue delay (not low drop rate); TCP variants and network environments independent. The problems of SDN and other solutions had been mentioned in this dissertation before the researcher involves machine learning to address the characteristics of the universe AQM by proposing Smart-AQM. The proposed smart AQM (RLAQM) achieved the generalization requirements.