Transmission Control Protocol (TCP) has been the workhorse of the Internet ever since its inception. The success of the Internet, infact, can be partly attributed to the congestion control mechanisms implemented in TCP. Though the scale of the Internet and its usage increased exponentially in recent past, TCP has evolved to keep up with the changing network conditions and has proven to be scalable and robust. However, the performance of TCP in Data Center Networks has been a major concern recently because it leads to impairments such as TCP Incast, TCP Outcast, Queue build-up and Buffer pressure. With cloud computing becoming an important part of the foreseeable future, it has become extremely important to enhance the performance of TCP in Data Center Networks and overcome these impairments. In this paper, we describe the above mentioned impairments in brief and then compare the TCP variants proposed so far to ovecome these impairments in Data Center Networks. The advantages and shortcomings of every TCP variant are highlighted with respect to its efficacy and the deployment complexity. A few open issues related to TCP's performance in Data Center Networks are also discussed.

No abstract

This paper presents the implementation and validation of PI 2 Active Queue Management (AQM) algorithm in ns-3. PI 2 provides an alternate design and implementation to Proportional Integral controller Enhanced (PIE) algorithm without affecting the performance benefits it provides in tackling the problem of bufferbloat. Bufferbloat is a situation arising due to the presence of large unmanaged buffers in the network. It results in increased latency and therefore, degrades the performance of delay-sensitive traffic. PIE algorithm tries to minimize the queuing delay by auto-tuning its control parameters. However, with PI 2 , this auto-tuning can be replaced by just squaring the packet drop probability. In this paper, we implement a model for PI 2 in ns-3 and verify its correctness by comparing the results obtained from it to those obtained from the PIE model in ns-3. The results indicate that PI 2 offers a simple design and achieves similar responsiveness and stability, and in some cases, better than PIE.

No abstract