Service based load balance mechanism using software-defined networks / Ahmed Abdelaziz Abdelltif Osman

Ahmed Abdelaziz, Abdelltif Osman (2017) Service based load balance mechanism using software-defined networks / Ahmed Abdelaziz Abdelltif Osman. PhD thesis, University of Malaya.

	PDF (The Candidate's Agreement) Restricted to Repository staff only Download (1761Kb)
Preview	PDF (Thesis PhD) Download (3952Kb) | Preview

Abstract

The proliferation of servers on the Internet has led to the emergence of server load balance as an important service in cloud aiming to optimize resource usage, maximize throughput and minimize the response time. Server load balance is the process and technology that distributes incoming requests among several servers in order tominimize the response time and maximize the utilization of servers. The existing schemes of the load balance (dynamic or static) do not consider the service types as well as the size of the request. Besides, these schemes are implemented either in dedicated hardware devices called load-balancer or built intothe Operating System (OS) such as Linux Virtual Server (LVS). It is difficult to customize the built-in LB scheme during runtime. Additionally, load balancer experiences problems due to the same scheme being used for different type of services. In the cloud, most Service Providers (SP) host various kinds of services that require different load balancing schemes. This requires theinstallation of additional load-balancers for each service or a manual reconfiguration of the device to handle the new services. Such operation is time-consuming and expensive(Marc Koerner & Kao, 2012). To address aforementioned problems, we proposed a service based load balance (SBLB) mechanism using Software-Defined Networks (SDN). The SDN controller is leveraged to provide online flow classification. The proposed mechanism is evaluated using benchmarking experiments and validated using a statisticalmodel. We investigate the impact of the different type of requests (compute and data) on SBLB mechanism in homogeneous and heterogeneous environments. The results demonstrate that SBLB can provide faster response time, higher throughput as compared to the other load balance solutions. For example, SBLB mechanism can reduce the average response time (ART) up to 5% and reply time (RT) up to 3% as compared to HAproxy load balance in the homogeneous environment. In the heterogeneous environment, SBLB mechanism demonstrates 51% decrease in average response time and 47% decrease in reply time as compared to theHAproxy load balancer. Besides, our proposed mechanism also outperforms round robin algorithm in both environments. SBLB shows 7% increases in request per second (RPS) in the homogeneous environments and 20% increases in request per second in the heterogeneous environment as compare to aRound-Robin algorithm.

Actions (For repository staff only : Login required)