Generation of synthetic workloads for distributed real-time computing systems.

The performance of a computing system depends on the workload it is executing. When evaluating a computer system, the system should be measured while executing a workload with known characteristics. Otherwise, the performance of the system can only be analyzed after the workload is characterized. An ideal tool for this situation is a synthetic workload. A synthetic workload (SW) is a set of artificial programs that serve as the workload for a system during performance measurement experiments. The programs are parameterized such that the user controls their structure and behavior and, thus, the resource demands they place upon the system. The SW can be made to produce resource demands that are representative of an actual workload, or it may be used to produce specific conditions that will exercise a particular subsystem under study. In this dissertation, we study the design of SWs and develop tools for generating SWs for distributed real-time systems. Since the SW is to be representative of an actual application, we structure it like an actual application. To do this, we have designed the SW based on a model of the structure of the software that composes real-time workloads. We describe a synthetic workload generator (SWG) which can produce large, distributed SWs from a compact description written using our SW specification language, SWSL. SWSL contains features to specify SWs for large systems and to support the process of conducting experiments. We have also developed a generic SW driver that controls the distributed SW as it executes on the target system. We demonstrate that the SW is capable of producing representative SWs, and that it may be used for the experimental evaluation of a distributed real-time system. These demonstrations have been performed on two different real-time computing systems, thus showing the generality of the SW design.PhDApplied SciencesComputer scienceUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/129002/2/9308362.pd