Congestion Control for TCP in Data-Center Networks

The several corresponding servers may send the data concurrently in a circle manner to a specific receiver. For example, data center applications such as Map-Reduce and Search applications are using many-to-one TCP traffic pattern i.e. parallel data delivery. While TCP connection is slow down the concurrent data deliver. The data delivery delay ensured the packet loss and time out in the TCP connection namely called TCP incast congestion. This congestion may reduce the performance of TCP connection such as increase the waiting time and increase bandwidth utilization. The proposed system is implemented for avoid the ICTCP congestion with assist of receive(r)-window-based congestion control algorithm. Initially, the proposed algorithm shared the available bandwidth namely called quota for all TCP connections. Next, assign the control interval to every TCP connections using Round Trip Time. The control interval will be 2*RTT for every TCP connections. Finally, adjust (increase/decrease) receive-widow size based on two threshold values, available quota, Round Trip Time, current TCP connection throughput in receiver side and expected throughput of the TCP connection. If ratio of difference between measured throughput and expected throughput is less than one expected throughput increase the receive-window-size. Otherwise, the ratio is large decrease the receive window. Further, improve our proposed system that converted many-to-many traffic congestion into many-to-many traffic congestion Keywords— Data-Center ,incast congestion, Round Trip Time ,TCP.