Keywords: cloud computing, job scheduling, ant colony optimization, genetic algorithms.

Scientists and engineers are more and more faced with the need of computational power to satisfy the ever increasing resource intensive nature of their experiments. Specifically, parameter sweep experiments (PSEs) are a class of experiments that enable independent data analysis over large parameter spaces. PSEs allow scientists to perform simulations by running the same scientific code with different input data, which typically results in many CPU-intensive jobs [1]. Clouds offer many technical and economic advantages over other platforms and combine customization of virtual machines (VM), scalability and resource sharing. The use of virtualization in particular has proved to deliver many useful benefits for scientific applications.

Within a cloud, the VMs are distributed among different physical resources or consolidated to the same machine to increase their utilization. To perform this, correctly scheduling the processing units on a cloud is an important issue and it is necessary to develop efficient scheduling strategies to appropriately allocate the VMs in physical resources. Scheduling here refers to the way VMs are allocated to execute on the available physical resources, since there are typically many more VMs running than physical resources.

A cloud scheduler, based on ant colony optimization (ACO), the most popular swarm intelligence technique, is descibed to allocate VMs to the physical resources belonging to a cloud. The aim of this paper is to experiment with the scheduler in dynamic (non-batch) scheduling scenarios in which multiple users connect to the cloud at different times to execute their PSEs. The main performance metrics to study are the number of serviced users (or throughput) by the cloud and the number of executed jobs per time unit. Another contribution of this proposal is the study of our scheduler together with an exponential back-off strategy to retry the allocation of failing VMs that aims at servicing as many users as possible. Comparisons performed based on real PSE job data [1] and alternative cloud schedulers -including random, a scheduler based on genetic algorithms [3] and min-min- suggest that our scheduler allows for a fair assignment of VMs and delivers competitive performance with respect to the number of executed jobs per user. Experiments were carried out using CloudSim [4], a cloud simulator that is widely employed for assessing cloud schedulers.

1

C. García Garino, M. Ribero Vairo, S. Andía Fagés, A. Mirasso, J.P. Ponthot, "Numerical simulation of finite strain viscoplastic problems", Journal of Computational and Applied Mathematics, 2012, In press.

2

C. García Garino, C. Mateos, E. Pacini, "Job scheduling of parametric computational mechanics studies on cloud computing infrastructures", International Advanced Research Workshop on High Performance Computing, Grid and Clouds. Cetraro (Italy), June 2012.

3

L. Agostinho, G. Feliciano, L. Olivi, E. Cardozo, E. Guimaraes, "A Bio-inspired Approach to Provisioning of Virtual Resources in Federated Clouds", in Ninth International Conference on Dependable, Autonomic and Secure Computing (DASC), DASC 11, pages 598-604. IEEE Computer Society, Washington, DC, USA, 12-14 December 2011.

4

R. Calheiros, R. Ranjan, A. Beloglazov, C. De Rose, R. Buyya, "CloudSim: A Toolkit for Modeling and Simulation of Cloud Computing Environments and Evaluation of Resource Provisioning Algorithms", Software: Practice & Experience, 41(1): 23-50, 2011.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £40 +P&P)