Systems with Heterogeneous Workloads

Abstract

Abstract Today’s workloads consist of applications having very different characteristics in terms of service requirements, performance objectives, and arrival processes. In this chapter we consider models executing this type of workloads, referred to as multiclass workloads , consisting of customers having significantly different service requests. In this case, to achieve the necessary accuracy in the performance forecast, it is required to model the workload as consisting of multiple class of customers and to compute the performance estimates for each class. A phenomenon characterizing multiclass models , is the bottleneck switch : the congested resource, the bottleneck , that determine the performance of the global system, can dynamically change depending on the modifications in the mix of applications in execution. When the bottleneck migrates from one resource to another, performance may change abruptly! The first case study described in this chapter is an example purposely designed to emphasize the impact of the bottleneck switch on the performance of the system. The second case study concerns the optimization of a data center with focus on the bottleneck identification and load balancing.