Performance evaluation of multipath TCP congestion control algorithms رسالة ماجستير

Abstract

The congestion control (CC) of the network has a considerable importance over all network applications. It is proved that Internet Service Providers (ISPs) need to develop a better understanding of modern concepts and strategies of the CC. For instance, whenever network design complexity increases the variety of congestion control algorithms (CCA) designs rises. These algorithms are important to enhance the network performance. The Internet Engineering Task Force (IETF) developed Multipath TCP CC. It enables the distribution of a single data stream across many paths. This strategy can improve the network's availability and bandwidth. It can also lead to efficient network usage and high utilization without being aggressive with competing flows. The behavior of a few of the best multi-path congestion control algorithms (MPTCP CCA) was investigated in this research. The purpose of this research is to describe the optimal performance of MPTCP CCA in a variety of situations, followed by the optimal implementation of the chosen algorithm. The optimal algorithm must perform better than single-path TCP in terms of throughput and fairness, and it must be capable of being used in place of TCP. Furthermore, we demonstrate that some multipath CCA might be damaging to the network's overall performance. Depending on a deep analysis of literature review, a performance analysis scenarios were built to investigate which is MPTCP CCA design has a better behavior on most situations. The study finding hypotheses were tested by using six MPTCP CCA in many situations. The study findings prove that none of the considered multipath algorithms is ideal. The MPTCP CCA is widely applicable and can be used to determine the behavior of a multipath variation of an improved CCA. We have demonstrated that MP transmission through MPTCP provides a significant performance improvement in certain scenarios. Furthermore, we found that the cubic CCA outperforms MPTCP OLIA, MPTCP BALIA, MPTCP, wVegas and regular TCP in all circumstances except when MPTCP is used without competing with other algorithms. Thus, even if we use many ISPs in our scenario, there will always be some costs associated with using MP techniques. However, the total performance benefits associated with MP transport remains significant. Each of these methods is implemented and evaluated on a Linux environment. Additionally, we evaluate it for multi-homed servers. In this study, the researchers used throughput as main metric, because it is a clear goal of most CCA mechanisms to maximize throughput, throughput and goodput are sometimes distinguished from one another. While throughput is the link utilization or flow rate in bytes per second, goodput is the subset of throughput made up of useful traffic and is also measured in bytes per second. The increasing throughput is important in a variety of environments, including overloaded and underused networks as well as long-lived and short-lived flows. For instance, throughput has been assessed in terms of the transfer times for connections with a variety of transfer sizes in order to evaluate Quick-Start, a proposal to enable flows to start up more quickly than slow-start. The researchers used ifstate tool to calculate the throughput in this study, the ifstat tool displays statistics for network interfaces. The interface maintains copies of the previous information shown in history files. It shows the difference between the most recent and current calls by default. Except the fourth scenario, MPTCP Cubic outperforms all other algorithms, resulting in an average throughput of 1908 (kb/s) in the first scenario, up 43% from the next-closest algorithm. In the second case, MPTCP Cubic performs better than every other algorithm and achieves a throughput of 1544 (kb/s), an increase of 18.6% over the next-closest algorithm. The third scenario experienced the same results, but MPTCP Cubic average throughput was slightly higher than the closest algorithm at 1247 (kb/s), an increase of just 14%. In the fourth scenario, all MPTCP algorithms outperform the cubic, and the MPTCP Cubic throughput was 60% less than the best while producing results that were similar to those of the regular algorithm.