Mechanism for reliable low latency communications in Computing Clusters
Saibal Kumar Ghosh, Dharma Prakash Agrawal · Transactions on Networks and Communications · 2015
Recent increases in the demands for computing power have given rise to the prevalence of distributed computing. Computationally complex problems are broken down into smaller chunks and are distributed to computing nodes that perform the computation simultaneously. The nodes may exchange information as peers and their combined result is the final outcome of the computation. Multiple computers need a mechanism to communicate and exchange information in order to harness collective computing power. Traditionally, the Transmission Control Protocol (TCP) has been used to exchange information between computers. However, additional overhead, generally associated with TCP has been considered as a serious drawback for any rapid data exchange. The User Datagram Protocol (UDP) alleviates the overheads of TCP but provides no reliability for the data transfer. In this paper, we introduce an enhanced UDP mechanism to exchange data of limited size reliably and without any associated overhead of TCP.