Improvement of fault tolerance in cloud data center networks

Abstract

Data center is disposed to failures due to the large number of devices used for the interconnections and communications. To recover from common failures with maintained adequate performance means that the design of the network must be firm; so, as the switches and links failure rates increase, such network is expected to exhibit fault tolerance and graceful performance degradation. This means that to achieve fault tolerant, and reliable performance, there must be provisions that will tolerate failures of devices. In the quest to measure-up with challenges in data center networks, several influential network architectures were designed, for example Fat Tree (FT), DCell and BCube. Meanwhile, Fat Tree, which is said to be widely used in data center network design because of its inherent fault tolerant capability, has not been able to resolve completely the fault tolerance issues in cloud data center network.
In line with this, we proposed an improved version of Fat Tree topology that is made up of the combination of a variant of Fat Tree topology called Ƶ-FT and its reversed form. We simply call this Hybrid or a Reversed Hybrid Fat Tree (𝐻_ℎ,𝑚). With this type of design, we were able to harness the full potentials of Fat Tree topology that gave it edge over other network architectures for the construction of cloud data center network. Our proposed Hybrid FT (H_h,m), because of its unique design that eliminates the deterministic routing (from root switch to the server) as seen in other variants of Fat Tree, was able to outperform the Single FT architectures that were compared with it.
Several simulations were carried out in Riverbed simulation tool and evaluated using NS-3 simulation tool, based on different packet sizes and traffic patterns. The results of these simulations show that our proposed Hybrid FTs with less links, still perform better than the Single FTs using the same number switches and servers. With our proposed design, we achieved greater percentage of received packets, minimum delay, graceful performance degradation, less percentage of packet loss, congestion control and fault tolerant.

Date of Award	Dec 2018
Original language	English
Awarding Institution	University of Portsmouth
Supervisor	Athanasios Paraskelidis (Supervisor) & Mo Adda (Supervisor)