An Enhanced SLA Based Framework for Bulk Provisioning In Telecommunication Networks

Abstract: High performance computing is possible through cluster computing where load is shared by a pool of networked servers. The cluster computing can serve jobs which are memory and CPU-intensive in batch processing model. Existing systems focused on balancing CPU loads ignoring page faults and I/O operations. Recently Xiao et al. explored policies to share loads that also consider memory usage in addition to load balancing in clusters. The policies can withstand predictable and unpredictable bursts in demands. In this paper we implement those policies and perform trace-driven simulations. The empirical results revealed that the proposed system is capable of capturing dynamic memory access patterns in the presence of uncertain workloads. Thus the policies are able to improve the performance in terms of execution of jobs by optimally utilizing memory and CPU resources.

Keywords: Cluster computing, load balancing, uncertain workloads, trace-driven simulations