Fault tolerance and graceful performance degradation in cloud data center networks

In relation to network performance, graceful degradation amidst increase in the failure of network devices expects a reduction in performance in a way that a dramatic fall in throughput will not be noticed. To achieve a relevant graceful performance degradation especially in a cloud data center networks means the design must be fault tolerant. A fault tolerant data center network should be able to provide alternative paths from source to destination during failures so that there will not be abrupt fall in performance. But this is not the case because of the growth in the use of internet-based applications, big data, and internet of things; leading to several ongoing researches to find the best suitable design that could help alleviate the poor fault tolerance and graceful performance degradation in cloud data center. Fat trees (FT) interconnections have been the most popular topologies used in data centers due to their path diversities and good fault tolerance characteristics.

In this paper, we propose a Reversed Hybrid architecture derived from the more generalized fat tree structure, Z-fat tree and compare it to a fat tree network with the same amount of resources for client server communication patterns such as HTTP and EMAIL application in a cloud data center. The results with faulty links show that our proposed Reversed Hybrid outperform the fat tree. We conclude based on the level of graceful performance degradation achieved that fault tolerance in data center cannot only be realized by adding extra hardware to the network, rather bespoke design plays a greater role.