Processor scheduling policies can be broadly divided into space-sharing and time-sharing policies. Space-sharing policies partition system processors and each partition is allocated exclusively to a job. In time-sharing policies, processors are temporally shared by jobs (e.g., in a round robin fashion). Space-sharing policies can be either static (processor allocation remains constant during the lifetime of a job) or dynamic (processor allocation changes in response to changes in job parallelism). Equipartition is a dynamic space-sharing policy that has been proposed and studied extensively. Among the time-sharing policies, job-based round robin policy (RRJob) has been shown to be a very good policy. Performance analysis of these two policies suggests that Equipartition policy performs well at low to moderate system loads and is extremely sensitive to system overheads and variance in service demand of jobs. RRJob performs better when there is a high variance in service demand and at hi...

Various processor allocation policies for multiprogrammed multiprocessor systems have been proposed in the literature. The focus of this paper is on the class of non-preemptive policies, since they represent a viable choice for implementation in actual systems. Several allocation policies are evaluated from the perspectives of both performance and applicability. A uniform comparison of policies is given based on the parameters required for implementation. A comparison of the several policies is conducted. The comparative evaluation demonstrates the conditions under which each policy performs well. Based on the comparison, an evaluation criterion for non-preemptive allocation policies based on the envelope of non-preemptive static equipartitioning policies is proposed. The criterion suggests a guide for the design of new non-preemptive policies and is useful in assessing the performance of existing policies. Limited theoretical justification of the evaluation criterion is given. 1 Intro...

In this paper a methodology for the evaluation of non-preemptive adaptive policies for multiprogramming multiprocessor systems is proposed. An evaluation criterion is derived, based on the systematic and consistent comparison of several allocation policies in a common framework. The comparison of the policies identifies the characteristic performance behavior of this type of policies. It suggests that equipartitioning strategies are an effective solution to the allocation problem when the workload components are statistically similar. The envelope of static equipartitioning policies is proposed as an evaluation criterion for non-preemptive adaptive policies. This criterion offers a guideline for the design of new policies and is useful in assessing the performance of existing ones. Analytical validation of the proposed criterion is presented for a simple case. 1 Introduction The increasing power of hardware coupled with inherent limitations on software parallelism suggest the use of m...

Processor scheduling policies for distributedmemory systems can be divided into space-sharing or timesharing policies. In space sharing, the set of processors in the system is partitioned and each partition is assigned for the exclusive use of a job. In time sharing policies, on the other hand, none of the processors is given exclusively to jobs; instead, several jobs share the processors (for example, in a round robin fashion). There are advantages and disadvantages associated with each type of policies. Typically, space-sharing polices are good at low to moderate system loads and when jobs parallelism do not vary much. However, at high system loads and widely varying job parallelism, time sharing policies provide a better performance. In this paper we propose a new policy that is based on a hierarchical organization that incorporates the merits of these two types of policies. The new policy is a hybrid policy that uses both space-sharing as well as time-sharing to achieve better perf...

This paper is dedicated to Prof. Horst Wettstein on the occasion of the 25th anniversary of his appointment. 1 Motivation

Processor scheduling policies on systems that run multiple applications simultaneously can be broadly divided into space-sharing and time-sharing policies. Space-sharing policies partition the system processors and each partition is allocated exclusively to a job. In timesharing policies, processors are temporally shared by jobs (e.g., in a round robin fashion). Space-sharing and timesharing policies have their advantages and disadvantages. It has also been suggested that a hybrid policy that combines space-sharing and time-sharing is beneficial in improving the overall performance. Processor scheduling has received considerable attention in the context of shared-memory multiprocessor systems but has not received as much attention in distributed-memory multicomputer systems. Furthermore, most previous research in this area has either used a simulation model or an analytical model in evaluating the performance of various policies. Very often these models neglect several practical aspect...

Processor Scheduling In A Distributed-Memory Computing Environment By Stephen W. Turner In recent years, the development of large-scale distributed-memory computers has given the user community unprecedented levels of computing power. In order to effectively use the available computing power, processor scheduling algorithms have been developed that allow many users to share distributed computing resources while obtaining the best possible job turnaround time. However, not all existing scheduling techniques take full advantage of available computing power. For example, in hypercubes, a cluster must normally be allocated as an entire subcube, which can result in high internal fragmentation, as well as poor job performance. Although the distributed workstation environment has recently become popular as a choice for a distributed-memory parallel computer, the problem of scheduling specifically for parallel job execution has not been well studied in this environment. In this thesis, we pr...

Introduction Even though distributed memory machines are the prevailing class for all highly parallel machines built in the last years, there are still architectural differences in hardware and system software. Older systems are designed and used as a large coprocessor to a `front-end-computer ' (e.g. T3D, [11]), modern systems allow stand-alone operation (e.g. T3E, [12]). In contrast to the coprocessor solutions, stand-alone operation requires an operating system on the parallel computer. Evolving a parallel operating system from a conventional microkernel based operating system (e.g. Mach, Windows-NT) seems to be a practible approach at first sight. But it is questionable if an operating system which was originally not designed with focus on parallel computing applications can provide adequate functional abstractions while still meeting the low latency, low overhead parallel application requirements. This causes many users to refuse operating system support and to rely on p