Modeling resource management in concurrent computing systems

Efficiently managing resources in very large concurrent computing systems can be viewed as a problem of spatial and temporal allocation of resources to running parallel programs. This basic parallel scheduling problem is considered and a framework is proposed for its definition and analysis. It is assumed that there are multiple parallel jobs executing on a large architecturally MIMD machine. Without loss of generality, each job is assumed to be data parallel, using as many virtual processors as necessary to exploit its inherent parallelism. The notion of virtual processor plays a unifying role in the conceptual design of a scheduler. The basic functions of any scheduler may be viewed as operations on the global set of job's virtual processors. In the context of the above framework, several open problems are identified, and in particular, the twin problems of spatial and temporal scheduling are addressed. Preliminary analysis points to the viability of horizontal spatial schedules and periodic temporal schedules.