Zahorjan J, McCann C. Processor scheduling in shared memory multiprocessors. Proceedings of the SIGMETRICS Conference on Measurement and Modeling of Computing Systems, May 1990; 214--225.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....reduce their demand for processors when the load on the system rises) from uncooperative applications (that do not) The strict division of processors into processor sets forces the uncooperative applications to compete against themselves instead of cooperative applications. Zahorjan and McCann [61] have conducted simulation studies of related scheduler policies that could be implemented in a Mach scheduling server. The scalability of Mach s processor allocation and control facility is an important issue because it is intended to adapt to future (larger) multiprocessor architectures. ....

John Zahorjan and Cathy McCann. Processor Scheduling in Shared Memory Multiprocessors. In SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May 1990.

....properties. We study the simple Equi partition algorithm, for which an equal number of processors is assigned to every job. The approach of Equi partition was first introduced to parallel scheduling by Tucker and Gupta as a process control policy [24] and later modified by Zahorjan and McCann [27] to dynamically adjust processor allocations as job requirements for processors change. This algorithm is known as dynamic equi partition (DEQ) We show that the Equi partition algorithm (which performs at most n preemptions) achieves a performance within 2 3 times the optimum schedule ....

....can be achieved with few reassignments of processors. Of course, much more research is required to make this theoretical understanding useful in a practical setting. In fact, performance in such systems has been already studied using simulation, experimental, and queuing theoretical approaches [2, 13, 17, 18, 24, 27, 1]. In this perspective, our research constitutes a theoretical confirmation of these efforts. 1.1 Modeling Job Execution In our model, all jobs arrive at time zero. That is, we adopt a batch job processing model. It would be more general to allow jobs to arrive at arbitrary times. However, this ....

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of the 1990.

....) This gives ffi ffiT , which is the length of time for EQUI s to complete the new fully parallelizable phase with work fi ffiT . 11 Open Problems The performance of Equi partition has been studied extensively using simulation, experimental, and queuing theoretical approaches [3, 13, 16, 17, 24, 28, 2]. Our research constitutes a theoretical confirmation of these efforts. The main open problem is to close the gaps between the lower bounds on the competitive ratio known for general non clairvoyant schedulers and those known for the specific schedulers Equi partition and Balance. This gap is ....

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of the 1990.

....n) However, the batch model assumes that the number of jobs n is at most the number of processors p because all these jobs arrive and are executed at once. Because our jobs arrive at arbitrary times, n is assumed to be much bigger then p. 28 13 Open Problems and queuing theoretical approaches [3, 13, 16, 17, 24, 28, 2]. Our research constitutes a theoretical confirmation of these efforts. The main open problem is to close the gaps between the lower bounds on the competitive ratio known for general non clairvoyant schedulers and those known for the specific schedulers Equi partition and Balance. This gap is ....

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of the 1990.

.... Allocation Because the efficiency of parallel jobs generally decreases as their processor allocation increases, it is necessary to decrease processor allocations to moldable jobs as the overall system load increases in order to avoid system saturation (see Sevcik [77] Zahorjan and McCann [97] found that allocating processors to evolving jobs according to their dynamic needs led to much better performance than either run to completion with a rigid allocation or round robin. For the overhead parameters they chose, round robin beat run to completion only at quite low system loads. ....

J. Zahorjan and C. McCann, "Processor scheduling in shared memory multiprocessors ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 214--225, May 1990.

....to run, the number of resources available in the system, the number of users available in the system, the response time of the operating system. We have not implemented this feature in the system, but it is definitely an important topic of research where several groups have been working [20, 3]. In Andorra I, slaves are subject to slave redeployments and masters are subject to master redeployments. Slave redeployments are (in order of cost) a) stay as slave; b) change to another team; c) become a master; d) be dismissed from its job (i.e. be released to the operating system to ....

John Zahorjan and Cathy McCann. Processor Scheduling in Shared Memory Multiprocessors. In 1990.

.... (in which the scheduler attempts to ensure that all processes of a given application run simultaneously) and processor partitioning (in which each application is given exclusive use of a subset of the processors of the machine for an extended period of time) A growing body of evidence [2, 5, 10, 15, 16] suggests that throughput is maximized by a processor partitioned environment in which each application runs exactly one process per processor. In such an environment, or in one that employs coscheduling, all the processes that share a given lock will always run simultaneously. Unfortunately, this ....

J. Zahorjan and C. McCann. Processor Scheduling in Shared Memory Multiprocessors. In PROC of the 1990.

....with EQS even when repartitioning is assumed to have zero cost. This is an impressive result since FB PWS commits to a processor allocation at job arrival time. Previous static allocation policies have generally not been competitive with dynamic policies such as EQS under zero repartitioning cost [13, 29, 19, 4]. They also show that FB PWS and FB ASP can substantially outperform EQS under an ad hoc model of repartitioning costs that is intended to illustrate the possible impact of repartitioning overheads on relative policy performance. In this paper, we further investigate the relative performance of ....

J. Zahorjan, C. McCann, Processor Scheduling in Shared Memory Multiprocessors. Proc.

....names) have been evaluated in many comparative studies of multiprocessor scheduling policies. The Dynamic Equipartition based rules have generally been found to yield excellent performance in shared memory environments (where the overhead of changing processor allocations can be kept low) [4, 11, 14, 15, 23] etc. 3.2 Adaptive Policies Adaptive policies assign a number of processors to each application when it is initiated, and the processors are freed only when the application completes. The number of processors allocated may depend on the current system load and on any available estimates of ....

....in non decreasing order of their demands. The pure adaptive policies use only the maximum parallelism of each application in doing processor allocation, while the above composite policies require also the total demand of each application. 3. 4 Differential Allocation Policies McCann and Zahorjan [15] introduce the Run To Completion policy, where each released processor is allocated to the waiting job for which the expected reduction in elapsed execution time is greatest, 7 ensuring that as many jobs as possible are activated. Wu [24] extends the Run To Completion policy with a Shortest ....

[Article contains additional citation context not shown here]

Cathy McCann and John Zahorjan. Processor Scheduling in Shared Memory Multiprocessors. Technical Report 89-09-17, Computer Science Department, University of Washington at Seattle, 1989.

.... 22, 25, 35] as is common in scientific engineering computing environments [6] Adaptive partitioning policies, where the number of processors allocated to a job is determined when jobs arrive and depart based on the current system state, have also been considered in a number of research studies [14, 42, 8, 21, 22, 33, 3, 25]. This approach tends to outperform its static counterparts by adapting partition sizes to the current load. However, the performance benefits of adaptive partitioning can be limited due to its inability to adjust scheduling decisions in response to subsequent workload changes. These potential ....

....limited due to its inability to adjust scheduling decisions in response to subsequent workload changes. These potential problems are alleviated under dynamic partitioning, where the size of the partition allocated to a job can be modified during its execution, at the expense of increased overhead [40, 4, 14, 42, 9, 16, 21, 22, 35]. The runtime costs of a dynamic partitioning policy are heavily dependent upon the parallel architecture and application workload under consideration. In uniform access, shared memory (UMA) systems, these overheads tend to be relatively small and thus the benefits of dynamic partitioning ....

[Article contains additional citation context not shown here]

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proc. ACM SIGMETRICS Conf., 214--225, 1990.

....27, 30, 34] which can be common in scientific and engineering computing environments. Adaptive partitioning policies, where the number of nodes allocated to a job is determined when jobs enter and leave based on the current system state, have also been considered in a number of research studies [21, 38, 16, 26, 27, 32, 7, 30]. This approach tends to outperform its static counterparts by adapting partition sizes to the current load. On the other hand, the performance benefits of adaptive partitioning can be limited due to its inability to adjust scheduling decisions in response to subsequent workload changes. These ....

....limited due to its inability to adjust scheduling decisions in response to subsequent workload changes. These potential problems are alleviated under dynamic partitioning, where the size of the partition allocated to a job can be modified during its execution, at the expense of increased overhead [37, 9, 21, 38, 17, 23, 26, 27, 34]. The relative runtime costs of a dynamic partitioning policy are heavily dependent upon the parallel architecture and application workload. In uniform access, shared memory systems, these overheads tend to be relatively small and thus the benefits of dynamic partitioning outweigh its associated ....

[Article contains additional citation context not shown here]

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of the ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 214--225, May 1990. 20

....using the remaining jobs and processors. Non integer values of t may be possible if the study is theoretical; otherwise some further decisions may have to be made to assign fractional parts of processors. Equipartition based algorithms have been shown to perform well in many studies [TG89, ZM90, CMV94, AS97] although their frequent repartitioning may result in high overheads. For this reason EQ is often used as an idealized baseline policy in studies of adaptive policies that attempt to approach its performance while limiting overhead costs. Since equipartitioning gives (roughly) equal ....

....applications exclusive use of a smaller number of processors is more efficient than uncoordinated timesharing or coscheduling. Even with multiple threads of the same application timesliced on a partition, applications did not experience as large of a slowdown as expected. Zahorjan and McCann [ZM90] compare an adaptive policy that used execution time functions (similar to Anastasiadis and Sevcik [AS97] as described above in Section 2.1.4) and a dynamic policy. The dynamic policy is actually closer to a variable partitioning policy, as the job model is evolving (see Section 2.1.3) allowing ....

[Article contains additional citation context not shown here]

J Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. Performance Evaluation Review, 18(1):214--225, May

....and in the timesharing band the priorities will be a precise indication of how much cpu time processes will accumulate relative to each other. The scheduler will implement a space sharing earnings based policy where the priority of a process will indicate the rate at which cpu time is accumulated [Zaho90][Poly94] Processes with high earnings will be scheduled ahead of processes with lower earnings and processes will be charged for their cpu usage when it occurs. The earnings value is an accurate indication of what a particular process s fair share of the machine, denominated in cpu microseconds, ....

J. Zahorjan, C. McCann, "Processor Scheduling in Shared Memory Multiprocessors", SIGMETRICS May 1990.

....hypercube and a MIN as example topologies for distributed and shared memory multiprocessors, respectively. It is shown that system performance can be significantly improved with this combined scheme compared to only 2 The definitions of static and dynamic partitioning are different from those in [34] [36] Here, they specify when and how a system is divided into subsystems. 3 Our understanding is that Intel Touchstone delta, Thinking Machines CM 5 and the future nCUBE 3 machines support (will support) multiprogramming. 5 multitasking. Also, all prior allocation policies are shown as ....

.... balancing between affinity and fairness have been explored in a shared memory environment [37] Overhead associated with the central processor manager and with the access contention for a single system queue is known as scheduler latency [40] Process control scheduling [33] two level scheduling [34] and pool based scheduling [38] policies attempt to reduce the scheduler latency by dividing the responsibility of the central processor manager into two parts. The central manager solely deals with the allocation of jobs to processors. Each job performs its own scheduling on the given set of ....

[Article contains additional citation context not shown here]

J.Zahorjan and C.McCann, "Processor Scheduling in Shared Memory Multiprocessors, " Proc. ACM SIGMETRICS Conf. Meas. and Mod. of Comp. Sys., pp.214-225, May 1990.

....properties. We study the simple Equi partition algorithm, for which an equal number of processors is assigned to every job. The approach of Equi partition was rst introduced to parallel scheduling by Tucker and Gupta as a process control policy [24] and later modi ed by Zahorjan and McCann [27] to dynamically adjust processor allocations as job requirements for processors change. This algorithm is known as dynamic equi partition (DEQ) We show that the Equi partition algorithm (which performs at most n preemptions) achieves a performance within 2 p 3 times the optimum schedule ....

....can be achieved with few reassignments of processors. Of course, much more research is required to make this theoretical understanding useful in a practical setting. In fact, performance in such systems has been already studied using simulation, experimental, and queuing theoretical approaches [2, 13, 17, 18, 24, 27, 1]. In this perspective, our research constitutes a theoretical con rmation of these e orts. 1.1 Modeling Job Execution In our model, all jobs arrive at time zero. That is, we adopt a batch job processing model. It would be more general to allow jobs to arrive at arbitrary times. However, this ....

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of the 1990 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 214-225, Boulder, Colorado, May 1990. 19

....sharing establishes dynamic partitions, where processors can freely migrate from one partition to another on line, according to the resource requirements of parallel programs and the operating system scheduling strategy. Although space sharing has proven to be beneficial for parallel programs [11, 19, 21], previous work has not addressed adequately the integration of space sharing in time sharing schedulers. As a consequence, most modern operating systems either do not support space sharing, or offer space sharing as an option on a per program basis, rather than as a concrete, transparent job ....

....with the exception of the SWIM workload with M = 4, where the difference is as much as 35 . The result is satisfactory although we could not find a good explanation for the SWIM workload. The corresponding average for irix mp, which uses a highly sophisticated earnings based scheduler for fairness [5, 15, 21] is 16 . Figure 2 illustrates the average turnaround times of the benchmarks in homogeneous workloads executed in closed system mode. The comparative results are analogous to the results from the experiments in open system mode. However, IRIX demonstrates a solid improvement over the results in ....

J. Zahorjan and C. McCann, Processor Scheduling in Shared Memory Multiprocessors, Proc. of the ACM SIGMETRICS'90 Conference on Measurement and Modeling of Computer Systems, pp. 214--225, Boulder, CO, 1990.

....each phase executes with linear speedup (up to the maximum degree of parallelism) multiple phases of execution with di#erent maximum degrees of parallelism can be combined to produce an overall model of sublinear speedup. In this paper we show that the dynamic equipartition (DEQ) policy [33] [38] produces mean completion times that are no more than 2 2 n 1 times the optimum for any set of n parallel jobs with one phase of parallelism, and that no policy can guarantee a better competitive ratio without a priori knowledge of job execution times. Although the competitive ratio turns ....

....phase of parallelism. This result provides a theoretical foundation for analyzing and understanding the performance of the DEQ policy, which, along with its various derivatives, has been shown to be superior to nonpreemptive algorithms in recent simulation and experimental studies [33] 17] [38], 16] 20] 21] The remainder of the paper is organized as follows. We complete section 1 with a further description of the problem and a discussion of related work. In section 2, we give a formal definition of the DEQ allocation policy. Then we establish a lower bound on the optimal total ....

[Article contains additional citation context not shown here]

J. Zahorjan and C. McCann, Processor scheduling in shared memory multiprocessors, in Proceedings of the 1990 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, ACM, New York, 1990, pp. 214--225.

....multiprocessor architecture, and special requirements of certain application workloads are among the factors affecting the effectiveness of the processor allocation strategies for multiprogramming parallel systems. Several processor allocation policies have been proposed in the literature [Oust82, MEB88, Sev89, TG89, PD89, LV90, DCDP90, ZM90, GST91, GTU91, MEB91, ZB91, MVZ93, SST93, RSDSC94, Sev94, CMV94, MZ94]. Each of these policies is designed to perform well under certain conditions. In this paper preemptive and non preemptive space sharing policies are considered. Under preemptive policies, parallel programs can be stopped during execution to allow for resource redistribution according to changing ....

J. Zahorjan, C. McCann, "Processor scheduling in shared memory multiprocessors," Proc. ACM SIGMETRICS, 1990, pp. 214-225.

....strategies for multiprogramming parallel systems. In this paper, effects of some of these factors on the performance of dynamic and adaptive space sharing policies are investigated empirically. Several dynamic and adaptive processor allocation policies have been discussed in the literature [MEB88, TG89, PD89, LV90, DCDP90, ZM90, GTU91, ZB91, MVZ93, SST93, RSDSC94, SEV94, CMV94, MZ94]. Performance analysis of dynamic space sharing scheduling policies has been largely based on simulation studies and Markovian analysis of small systems. For a simulation study to be accurate and realistic, detailed knowledge of various parameters of the system under consideration is necessary. ....

J. Zahorjan, C. McCann, "Processor scheduling in shared memory multiprocessors, " Proc. ACM SIGMETRICS, 1990, pp. 214-225. This article was processed using the L a T E X macro package with LLNCS style

....[23] The kernel then is forced to choose which of the first job s kernel threads, and thus implicitly which user level threads, to run on the remaining processors. The need to preempt processors from an address space also occurs due to variations in parallelism within jobs; Zahorjan and McCann [27] show that the dynamic reallocation of processors among address spaces in response to variations in parallelism is important to achieving high performance. While a kernel interface can be designed to allow the user level to influence which kernel threads are scheduled when the kernel has a choice ....

....for our prototype implementation; we briefly describe these, as well as some performance enhancements and debugging considerations, in the subsections that follow. 4. 1 Processor Allocation Policy The processor allocation policy we chose is similar to the dynamic policy of Zahorjan and McCann [27]. The policy space shares processors while respecting priorities and guaranteeing that no processor idles if there is work to do. Processors are divided evenly among the highest priority address spaces; if some address spaces do not need all of the processors in their share, those processors are ....

ZAHORJAN, J., AND MCCANN, C. Processor scheduling in shared memory multiprocessors. In Proceedings of the 1990 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems (Boulder, Colo., May 1990), pp. 214--225.

.... allocation to each runnable job in order to minimize average response time 1 Introduction Much of the work on scheduling policies for multiprogrammed multiprocessors has focused on how many processors to allocate to each runnable job without considering the memory requirements of those jobs [7, 14, 15, 6, 17, 18, 13, 8, 2, 9, 16]. In this paper we consider jobs whose memory requirements imply a lower bound on the amount of machine resource they can be allocated for execution. The interaction of processor scheduling and job memory usage has been considered in [12] However, they examined a paging environment, with the ....

.... of shared memory systems [15, 6] For parallel machines, however, space sharing, in which resources are divided equally by partitioning the processors among the jobs, has proven to be more effective than time sharing the machine by rotating all resources among the jobs in a round robin fashion [15, 17]. Among the benefits of spacesharing is that it interacts well with co scheduling, which requires that either all processes of a single parallel application be scheduled at once or that none be scheduled. Co scheduling has long been known to be critical to parallel machine scheduling [11] Based ....

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In Proceedings of ACM SIGMETRICS Conference, pages 214--225, May 1990.

No context found.

Zahorjan J, McCann C. Processor scheduling in shared memory multiprocessors. Proceedings of the SIGMETRICS Conference on Measurement and Modeling of Computing Systems, May 1990; 214--225.

No context found.

J. Zahorjan and C. McCann. Processor scheduling in shared memory multiprocessors. In ACM SIGMETRICS Conference on Measurement and Modelling of Computer Systems, pages 214--225, May 1990.

No context found.

Zahorjan, J. and McCann, C. Processor Scheduling in Shared Memory Multiprocessors. In Proceedings of the 1990 AGM SIGMETRIGS Conference on Measurement and Modeling of Computer Systems, pages 214-225, May 1990.

No context found.

J. Zahorjan and C. McCann. "Processor Scheduling in Shared Memory Multiprocessors". In Proceedings of the 1990.

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC