Mitigating Incast Congestion with LTTP for Many to One Communication in Data Centers

The Transmission Control Protocol (TCP) is an important transport layer protocol which provides reliable data delivery. In this scientific world, number of users of internet and flow of data increases day by day making reliable data transmission difficult. As a result, in many to one communications there arises a problem known as TCP Incast which causes a severe drop in throughput. This arises due to the overflow of output buffer of switch and TCP's low Retransmission Timeout as compared to Round Trip Time. Earlier solutions included controlling switch buffer or updating OS/hardware. In this paper, a new transport protocol is proposed by improving the UDP-based transmission with a simplified FEC scheme and CRC. Through ns2 based simulations it is observed that the newly designed protocol will not degrade the throughput in many to one communications when the number of servers is increased and reduces the processing time when compared with LTTP protocol.