Improved Adaptive Resource Scheduling Technique for Computational Lambda Grid Network.

SOURCE:

Faculty: Engineering
Department: Electronic And Computer Engineering

CONTRIBUTORS:

Ilo, S.F.
Inyiama, H. C.

ABSTRACT:

Abstract
A grid network that employs optical wavelength division multiplexing (WDM) technology and optical switches to interconnect computing resources with dynamically provisioned multi-gigabit rate bandwidth light-path is called a Lambda Grid. Data-intensive Grid applications require huge data transfers between grid computing nodes. Many data-intensive e-science applications like electronic very long Baseline Interfemetry (e-VLB) and Genomes – to – life (GTL) require aggregating several hundred Gigabytes of data files from distributed databases to computing resources (such as super computers) frequently in real time. Since data is aggregated at the time of computation, the time required to transfer the data over the network may be the main computational bottleneck. The problem of reserving bandwidth in a lambda grid has been studied extensively in the literature resulting in the proposition of a number of resource scheduling algorithms. Unfortunately, none of the existing lambda grid scheduling algorithms dynamically readjusts the scheduler to accommodate the actual amount of time that is required to transfer a file. This dissertation is focused on the investigation of the existing technique and proposed improved an adaptive resource scheduling technique to minimize the delay in the data aggregation task required by the computational and data intensive e-science application running on lambda grid network. Simulation was carried out using the digital model of the 24 node National Lambda Rail (NLR) lambda grid network topology created with Cisco packet tracer 7.1. Results obtained showed that the proposed algorithm achieved 14% and 30% Average Finish Time improvement over the Varying Bandwidth List Scheduling Algorithm (VBLS) and Virtual Finish (ViFi) algorithms respectively. The proposed technique achieved a substantial reduction in blocking probability. It achieved 15.4% and 23% improvement in blocking probability over the VBLS and ViFi algorithms respectively. Results obtained for the effect of connection duration on blocking probability showed improvement of 7.5% and 18% over the VBLS and VIFi algorithm respectively. At 7.55%, the proposed algorithm showed a very low job blocking rate and indicates 16% and 29% improvements compared to the VBLS and ViFi algorithms respectively. Reduction in the variation of the effectiveness of the algorithm with job size was found. It achieved 18% and 21% improvement over the VBSL and ViFi algorithms respectively. Results also indicated that the proposed algorithm gives a substantially low reservation delay as per impact of request of lambda arrival rate.