Brecht, T., Guha, K.: Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Performance Evaluation 27/28 (1996) 519--539  Home/Search   Document Details and Download   Summary   Related Articles   Check

....seems to have a competitive ratio of 1:48 . There are other classes of speedup functions that are definitely not fully parallelizable, because Gamma(p) p yet that do not fit our definition of strictly sublinear. Such jobs seem to occur in practice and are used in many simulations [2]. It is interesting, that the Motwani like lower bound for these is the same as that for fully parallelizable jobs. The key property is that these speedup functions are fully parallelizable when the job is allocated a small number of processors, e.g. or Gamma(fi ) b fifi 1) b fi fi) ....

....) 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 ....

T. Brecht and K. Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. Performance Evaluation, 27 & 28:519--539, Oct. 1996.

....seems to have a competitive ratio of 1:48 . There are other classes of speedup functions that are definitely not fully parallelizable, because Gamma(p) p yet that do not fit our definition of strictly sublinear. Such jobs seem to occur in practice and are used in many simulations [2]. It is interesting, that the Motwani like lower bound for these is the same as that for fully parallelizable jobs. The key property is that these speedup functions are fully parallelizable when the job is allocated a small number of processors, e.g. or Gamma(fi ) b fifi 1) b fi fi) ....

....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 ....

T. Brecht and K. Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. Performance Evaluation, 27 & 28:519--539, Oct. 1996.

....or the amount of work of the system. These parameters are either already known or can be directly measured by the scheduler. 2 of 15 However, many researchers have shown that using application characteristics such as the speedup or the average parallelism improve the performance of the scheduler [2][3] 12] In particular, Parsons and Sevcik [13] showed that, if applications executing in a system have different speedup curves, the knowledge of this behavior is useful to the scheduler, since it can assign more processors to those applications that will take advantage of them. The speedup is ....

T.B. Bretch and K. Guha, "Using parallel program characteristics in dynamic multiprocessor allocation policies". IEEE Performance Evaluation 27&28 (1996) 519-539. ftp://www.cs.yorku.ca/pub/brecht/Brecht_Guha.ps. York University. 15 of 15

.... (SUBSET) either either either migratable Cosched (other) Ous82] Round Robin [ZM90] no no no RRJob [MVZ93] FB ASP,FB PWS no pws no LSF MIG LSF MIG AD(SUBSET) either either either malleable Equi Dynamic Partition [TG89,MVZ93] no no no FOLD,EQUI [MZ94] no no no (not applicable) W E [BG96] yes yes no BUDDY,EPOCH [MZ95] no no yes MPA [PS96b,PS96a] no yes yes LSF MALLAD (SUBSET) either either either previously scheduled job s execution (i.e. a gap exists in the schedule containing enough processors for sufficient time) The disciplines that we present in this paper have been ....

Timothy B. Brecht and Kaushik Guha. Using parallel program characteristics in dynamic processor allocation policies. Performance Evaluation, 27&28:519--539, 1996.

....as the SGI Origin2000 [29] This work attacks the problem of the processor allocation in an execution environment where no knowledge of the application is available when applications are submitted. Many researchers have considered the use of application characteristics in processor scheduling [3][6] 17] 22] 23] 24] In these works, parallel applications are characterized by different parameters such as the maximum speedup, the average parallelism, or the size of the working set. Performing the processor allocation without taking into account these characteristics can result in a bad ....

....were calculated in two different ways. The first approach is that the user or system administrator performs several executions under different scenarios, such as the input data or the number of processors, and collects several measurements. A second approach, used in research environments [3][6] 11] 14] 15] 24] 27] defines a job model, characterizing the applications by a set of parameters, such as the average of parallelism or the speedup. This information is provided to the OS as an a priori input, to be taken into account in subsequent executions. 1. Those scheduling policies ....

[Article contains additional citation context not shown here]

T.B.Brecht,K.Guha."Using Parallel Program characteristics in dynamic processor allocation", Performance Evaluation, 27&28 (1996) 519-539

....et al. 3] Leutenegger and Vernon [12] Sevcik [22, 23] Ghosal et al. 9] Rosti et al. 21] and others have proposed using application characteristics such as speedup, average parallelism, and processor working set to improve the performance of static processor schedulers. More recently, Guha [10] has proposed that application characteristics such as efficiency and execution time can also be used profitably by dynamic processor schedulers. All of these studies, however, assume that accurate historical performance data is provided to the scheduler at job submission time. In contrast, we ....

K. Guha. Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Technical Report CS95 -03, Department of Computer Science, York University, May 1995.

....current implementation to find sufficient resources to meet the user s deadline. A second group of researchers has studied the use of parallel application characteristics by processor schedulers of multiprogrammed multiprocessor systems, typically with the goal of minimizing average response time [10, 11]. However, the results from these studies are not applicable in our case because they were focussed basically on the allocation of jobs in shared memory multiprocessors in which the computing resources are homogeneous and available during all the computation. Moreover, most of these studies assume ....

T. B. Brecht and K. Guha, "Using parallel program characteristics in dynamic processor allocation policies", Performance Evaluation, Vol. 27 and 28, pp. 519-539, 1996.

....and availability of the system resources at scheduling time. There are other works in the literature that have studied the use of parallel application characteristics by processor schedulers of multiprogrammed multiprocessor systems, typically with the goal of minimizing average response time [8]. The results from these studies are not directly applicable in our case because they were focussed on the allocation of jobs in shared memory multiprocessors without considering the problem of task scheduling within a fixed number of processors. However, their experimental results also confirm ....

T. B. Brecht and K. Guha, "Using parallel program characteristics in dynamic processor allocation policies", Performance Evaluation, Vol. 27 and 28, pp. 519-539, 1996.

....jobs on parallel machines. There has been considerable work on this topic in the past. A great deal of this work has been involved with the interesting question of how best to schedule jobs 1 when given perfect information about speedup, a fundamental attribute of parallel workloads (e.g. [21, 8, 3, 2], among many others) While it would be useful to understand how to best schedule a set of jobs given a priori information on their speedups, such information is difficult, if not impossible, to specify accurately in practice 1 because of the sensitivity of job performance to the input data set ....

T. B. Brecht and K. Guha. Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Technical report, Department of Computer Science, York University, in preparation.

....et. al [3] Leutenegger and Vernon [12] Sevcik [22, 23] Ghosal et al. 9] Rosti et al. 21] and others have proposed using application characteristics such as speedup, average parallelism, and processor working set to improve the performance of static processor schedulers. More recently, Guha [10] has proposed that application characteristics such as efficiency and execution time can also be used profitably by dynamic processor schedulers. All of these studies, however, assume that accurate historical performance data is provided to the scheduler at job submission time. In contrast, we ....

K. Guha. Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Technical Report CS-95-03, Department of Computer Science, York University, May 1995.

....loaded systems, the job would get the minimal number, which are then guaranteed to be kept busy. In an unloaded system, the job should get up to the maximal number if there is nothing else to do with them. In general, information about the total work and the efficiency of the job is beneficial [81]. The problem with all the above schemes is that they assume that precise information about the application characteristics is available, and that the application cooperates with the system to find the optimal partition size. Note, however, that this is the optimal size from the system s point of ....

T. B. Brecht and K. Guha, "Using parallel program characteristics in dynamic processor allocation policies". Performace Evaluation 27&28, pp. 519--539, Oct 1996.

....loaded systems, the job would get the minimal number, which are then guaranteed to be kept busy. In an unloaded system, the job should get up to the maximal number if there is nothing else to do with them. In general, information about the total work and the efficiency of the job is beneficial [44]. 25 Memory considerations All the above algorithms, whether based on application characteristics or on system load, share one major oversight: they ignore memory requirements. In shared memory systems, this can lead to overallocation of global memory. In distributed memory machines, an ....

T. B. Brecht and K. Guha, "Using parallel program characteristics in dynamic processor allocation policies". Performace Evaluation 27&28, pp. 519--539, Oct 1996.

....Chiang et. al [3] Leutenegger and Vernon [8] Sevcik [15, 16] Ghosal et al. 5] Rosti et al. 14] and others have proposed using application characteristics such as efficiency, speedup, and average parallelism to improve the performance of static processor schedulers. More recently, Brecht [2] has proposed that application characteristics such as efficiency and execution time can profitably be used by dynamic processor schedulers as well. All of these studies, however, assume the availability of accurate historical performance data, provided to the scheduler simultaneously with job ....

T. B. Brecht and K. Guha. Using Parallel Program Characteristics in Dynamic Multiprocessor Allocation Policies. ftp://www.cs.yorku.ca/pub/brecht/Brecht Guha.ps, May 1995.

....questions through three sets of measurements. First, we examine application speedup, and the sources of speedup loss. Our results confirm that there is considerable variation among jobs, and provide information that will support work on the use of speedup information in making scheduling decisions [13, 23]. Second, because it is at least burdensome, and perhaps impossible, to accurately collect and supply such information at job submission time, we look at the problem of estimating job speedup at runtime. We first demonstrate a technique for estimating instantaneous speedup at runtime that is ....

K. Guha. Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Technical Report CS-95-03, Department of Computer Science, York University, May 1995.

....use of application execution time predictions obtained from the Historical Profiler lead to improved performance. 1 Introduction Many theoretical and modeling based studies indicate that knowledge of the characteristics of parallel applications can improve the performance of scheduling algorithms [MEB90, PD89, GST91, MEB91, Wu93, PS95, AS97, BG96, PS96]. However, much less research has focused on practical ways of obtaining such application knowledge. Sevcik [Sev94] proposes simply having the user provide estimates of application characteristics. However, this is inconvenient for users and the accuracy of data is not assured. As a result, more ....

T.B. Brecht and K. Guha. Using parallel program characteristics in dynamic processor allocation policies. Performance Evaluation, 27(8):519-- 539, October 1996.

....a parallel program that can run on a range of cluster sizes. The allocation strategy is the part of the scheduler that chooses the cluster size for each malleable job. this model [10] 14] Brecht and Guha use a variation of this model that imposes an upper bound on the speedup of some jobs [1] [9]. One problem with this model is that the parameter fi has little semantic content. Thus, it is not clear how to use observations of a real program to find the value of fi or how to choose a distribution of values that describes a real workload. As a result, workload models based on Dowdy s ....

Kaushik Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. Technical Report CS-95-03, York University, May 1995.

....is a parallel program that can run on a range of cluster sizes. The allocation strategy is the part of the scheduler that chooses the cluster size for each malleable job. this model [10] 14] Brecht and Guha use a variation of this model that imposes an upper bound on the speedup of some jobs [1] [9] One problem with this model is that the parameter fi has little semantic content. Thus, it is not clear how to use observations of a real program to find the value of fi or how to choose a distribution of values that describes a real workload. As a result, workload models based on Dowdy s ....

....propose a model for efficiency of data parallel tasks; they use measurements of ScaLAPACK programs to validate this model. Many of the allocation strategies that have been proposed for malleable jobs assume that the scheduler knows the average parallelism of all jobs [16] 8] 15] 17] 3] 12] [1]. Thus all of these strategies require that the parallelism profile of the program be known, or that A (and maybe V ) can be calculated by other means. Our model may provide a way to derive these characteristics. 2 The model The design goal for our speedup model is to find a family of speedup ....

Timothy B. Brecht and Kaushik Guha. Using parallel program characteristics in dynamic processor allocation policies. In Proceedings of Performance '96, October 1996.

....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 ....

T. Brecht and K. Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. Performance Evaluation, 27 & 28:519--539, Oct. 1996.

....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 ....

T. Brecht and K. Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. Performance Evaluation, 27 & 28:519-539, Oct. 1996.

....scheduling model. This result for nonpreemptive algorithms has been subsequently improved and extended [34] 27] 18] A number of di#erent preemptive policies have been proposed and studied for scheduling parallel jobs in multiprocessors [33] 29] 38] 20] 28] 26] 23] 24] [2]. In particular, experimental and simulation studies have shown that the DEQ algorithm yields low mean completion times under a variety of workloads and is reported to possess desirable properties of a good scheduler [33] 17] 16] DEQ was first introduced to parallel scheduling by Tucker and ....

T. Brecht, and K. Guha, Using parallel program characteristics in dynamic processor allocation policies, Performance Evaluation, 27--28 (1996), pp. 519--539.

.... algorithms has been subsequently Preemptive Scheduling of Parallel Jobs on Multiprocessors 5 improved and extended [34] 27] 18] A number of different preemptive policies have been proposed and studied for scheduling parallel jobs in multiprocessors [33] 29] 38] 20] 28] 26] 23] 24] [2]. In particular, experimental and simulation studies have shown that the DEQ algorithm yields low mean completion times under a variety of workloads and is reported to possess desirable properties of a good scheduler [33] 17] 16] DEQ was first introduced to parallel scheduling by Tucker and ....

T. Brecht, and K. Guha, Using Parallel Program Characteristics in Dynamic Processor Allocation Policies, Performance Evaluation, Volume 27 and 28, October, 1996, pp. 519-- 539.

....when used with models of efficiency that consider jobs with limited parallelism and jobs that experience slowdown when too many processors are allocated. Note that more detailed and accurate models of efficiency, including a model that addresses both of these issues, are examined by Guha [12]. 3.2 The Workload Model We assume that the work a job has to execute, W i , is drawn from a Hypergeometric distribution with mean W = 1000 and coefficient of variation CW (1, 5, and 30 are considered) This is consistent with variations in service demands used in previous studies [5, 26] and ....

....= Gamma10 for different CW ; load = 90 CW R (ff = 0) R (ff = Gamma10) Impr 1 100.1 1.7 36.5 0.4 64 5 100.4 5.2 29.8 0.7 70 30 98.1 5.1 28.2 2. 5 71 These same experiments have also been conducted under the assumption that the work, W i , is not known a priori (see [12] for details) By keeping track of each job s accumulated execution time and using that as an estimate of a job s remaining work, influenced by studies performed by Leland and Ott on uniprocessor systems [17] a portion of this study was recently confirmed for modern workloads by Harchol Balter ....

[Article contains additional citation context not shown here]

K. Guha. Using parallel program characteristics in dynamic multiprocessor allocation policies. M.Sc. Thesis, Technical Report CS-95-03, York University, North York, Ontario, May 1995.

No context found.

Brecht, T., Guha, K.: Using Parallel Program Characteristics in Dynamic Processor Allocation Policies. Performance Evaluation 27/28 (1996) 519--539

No context found.

T. Brecht and K. Guha, "Using Parallel Program Characteristics in Dynamic Processor Allocation Policies," Performance Evaluation, vol. 27/28, no. 4, pp. 519--539, 1996. [Online]. Available: citeseer.nj.nec. com/brecht96using.html

No context found.

T. B. Brecht, K. Guha. "Using Parallel Program characteristics in dynamic processor allocation", Performance Evaluation, 27&28, pp. 519-539, 1996.

First 50 documents

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