Indian Journal of Science and Technology

Abstract

Thousands of CPUs and storages are required for the analysis of scientific peta-byte scale data. The network interfaces of these systems are connected on a data center network for cluster computing and Grid computing for high throughput. Therefore, routing configuration of these network interfaces is usually set by internal routing protocols and set by having the only single gateway as a last resort or a default routing path. This style of routing configuration is the most typical configuration of routing protocols for the network of the legacy data center. And, as the volume of data is getting dramatically increasing due to the popularization of big data services, data center rapidly grow to have hundreds of thousands servers. Therefore, real routing path frequently changes by the fault of network devices or systems in large-scale data center environment even though data center networks works well apparently. The inefficiency of this kind of routing scheme caused by the instability of the internal routing protocol begins to appear as a considerable problem in big data center network management lately. In this paper, we showed the serious decrease of the throughput due to very short flapping time of the internal routing protocol by NS2 simulation.
Keywords: Default Ip Routing, Domestic Route Flapping, Large-Scale Data, Network Throughput