Loading Please Wait...

Image

ISSN: 2581-3560

RJSCES - Volume 3(Issue 1) 2018

Survey on Congestion Control Frameworks in Data Center Networks

Author: K. Arun Selvi

Other Authors: Dr. K. Kumar

Category: Engineering


Abstract:

Cloud datacenters which implements scale out distributed storage and computing frameworks such as Hadoop, Map Reduce, HDFS, Cassandra, etc. commonly have many-to-one communication pattern called incast. This pattern relies heavily on TCP and begins when a singular parent server places a request to a cluster of nodes simultaneously, leading to microburst of many machines concurrently sending TCP data streams to one machine (many-to-one). The synchronicity of such incast traffic with other throughput-demanding elastic traffic flows in Data Center Networks (DCNs) causes severe performance degradation in applications such as web search, maps, social networks, data warehousing and analytics. It also brings poor Quality of Service (QoS) and enterprise revenue loss. Over the past few years, many promising proposals have been put to address this incast problem. This paper made an in-depth survey on these proposals and summarizes the principles, merits and applications of the main proposals for solving the TCP incast congestion problem.