Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job scheduling scheme for pure space sharing among rigid jobs. In Proceedings of 4th Workshop on Job Scheduling Strategies For Parallel Processing, March 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....for both clusters of workstations and specialized parallel machines (e.g. massively parallel processors) using traditional performance metrics. Representative examples include [9] which evaluates three traditional batch scheduling algorithms (FIFO, SJF, LJF) for clusters of workstations, [1] which evaluates batch schedulers for the NEC Cenju 3 supercomputer, and [2] which evaluates batch systems for the SGI Origin 2000 parallel machine. In all cases, traditional performance metrics such as average waiting time and average response time were used as the basis for system evaluation. ....

Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job scheduling scheme for pure space sharing among rigid jobs. In Proceedings of 4th Workshop on Job Scheduling Strategies For Parallel Processing, March 1998.

....a good metric for a single job, many researchers have used the arithmetic mean mean( xxxx to combine the turn1 Turn around time is also referred to as service time or response time. 57 around times of all jobs in the workload into a metric for the supercomputer scheduler [1] [44] 64] The mean has the advantage of being easily understood and widely used to combine multiple experiments into a single value. For example, the arithmetic mean is a sensible choice for establishing the true value of physical measurements because measurement errors are reduced by ....

....preemption (i.e. preemption within the threads in a processor) and (iii) preemption over parallel jobs as a whole (a.k.a. gang scheduling) Most supercomputers currently run jobs until completion. This makes nopreemptive policies of great interest for researchers in supercomputing scheduling [1] [2] 24] 32] 43] 66] 67] 69] 82] 90] Local preemption seems to be of little value for supercomputer scheduling due to their negative impact on the performance of tightly coupled parallel jobs [38] Gang scheduling has been shown to improve performance under a variety of scenarios [42] ....

[Article contains additional citation context not shown here]

Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1459, 1988.

....our research scenario. Mean Turn Around Time Since turn around time provides a good metric for a single job, many researchers have used the arithmetic mean mean( xxxx to combine the turn around times of all jobs in the workload into a metric for the supercomputer scheduler [1] [15] The mean has the advantage of being easily understood and widely used to combine multiple experiments into a single value. However, in a typical supercomputer workload, jobs differ widely in execution time 12 (and thus in turn around time) 11] 15] The problem this causes is that mean ....

....the high cost of task migration, and (iii) the need to simultaneously make available all processors allocated to a job. In current practice, supercomputer schedulers accept rigid requests [17] 20] 22] 27] and thus much of the research available in the literature assume jobs to be rigid, e.g. [1] [2] 14] 21] 33] Closer to our own work, there has been studies on processor allocation [3] 8] 10] 12] 16] 19] 23] 25] 28] 29] 30] 31] Processor allocation consists of enabling the supercomputer scheduler to select how many processors to allocate to a parallel job based on ....

K. Aida, H. Kasahara, and S. Narita. "Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs". Job Scheduling Strategies for Parallel Processing, Springer-Verlag LNCS vol. 1459, 1988.

....backfilling is that in EASY it is guaranteed that the first job in the waiting queue will not be delayed while in conservative backfilling all jobs are guaranteed not to be delayed. Both algorithms achieve similar performance [2] while preventing the starvation problem. Several studies [1, 2, 10] showed that performance improvements achieved by backfilling as compared to first come first served policy are significant. The simulation study [1] showed that backfilling performs close, but slightly worse, to the queue searching policies that do not consider job running times and have ....

....are guaranteed not to be delayed. Both algorithms achieve similar performance [2] while preventing the starvation problem. Several studies [1, 2, 10] showed that performance improvements achieved by backfilling as compared to first come first served policy are significant. The simulation study [1] showed that backfilling performs close, but slightly worse, to the queue searching policies that do not consider job running times and have starvation problem. There is a single study [11] that uses queue sorting in the context of backfilling. It shows that mean job slowdowns are much better ....

[Article contains additional citation context not shown here]

K. Aida, H. Kasahara, and S. Narita, "Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs," in .lob Scheduling Strategies for Parallel Processing, vol. 1291, Lecture Notes in Computer Science, D. G. Feitelson and L. Rudolph, Eds.: Springer-Verlag, 1998.

....from different clusters) Because of its synchrony, the performance of a job performing such an algorithm is determined by its slowest task. 5 Related work The problem of scheduling rigid jobs by pure space sharing in a multiprocessor system has been studied extensively; for instance, see [1]. Whereas we approach the problem of the maximal utilization from a more theoretical perspective, in [14] a study of the utilizations as observed in existing supercomputing installations is presented. Experience with a large range of machines over more than a decade shows that employing FCFS ....

K.Aida, H.Kasahara and S.Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science 1459, pages 98121. Springer-Verlag, 1998.

....the high cost of task migration, and (iii) the need to simultaneously make available all processors allocated to a job. In current practice, supercomputer schedulers accept rigid requests [20] 23] 25] 28] and thus much of the research available in the literature assume jobs to be rigid, e.g. [1] [2] 16] 19] 24] 33] 35] Closer to our own work, there has been studies on processor allocation [3] 5] 12] 14] 18] 29] 30] 31] 32] Processor allocation consists of selecting how many processors to allocate to a parallel job based on information about the characteristics of the job ....

....today are moldable. In fact, in a survey conducted among supercomputer users, 98 of the respondents said that their jobs were moldable [7] SA s goal is to improve job j s turn around time by selecting the request to be sent to the supercomputer. Users provide SA with a set of requests r = r [1] , r [v] that can be used to submit a job j. Recall that a request r [i] n [i] tr [i] where n [i] is the partition size and tr [i] is time requested for the job, when running with n [i] processors. SA simulates the submission of all requests in r, and then selects the ....

[Article contains additional citation context not shown here]

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Feitelson and Rudolph (Eds.), Springer-Verlag Lecture Notes in Computer Science, vol. 1459, 1998. 27

....size n, requested time tr, request accuracy a) is known. Let v be number of requests that are available for SA to choose from. The moldability model thus uses n, tr, and te to produce the v requests that can be used to submit job j. That is, the moldability model produces the v tuples n = n [1] . n [v] tr = tr [1] tr [v] a = a [1] a [v] that describe v requests for job j. A moldable job is, by definition, a parallel job that can use partitions of various sizes to run. Note, however, that this definition does not mean that a moldable job can run ....

....request accuracy a) is known. Let v be number of requests that are available for SA to choose from. The moldability model thus uses n, tr, and te to produce the v requests that can be used to submit job j. That is, the moldability model produces the v tuples n = n [1] n [v] tr = tr [1] , tr [v] a = a [1] a [v] that describe v requests for job j. A moldable job is, by definition, a parallel job that can use partitions of various sizes to run. Note, however, that this definition does not mean that a moldable job can run over partitions of any size. ....

[Article contains additional citation context not shown here]

Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1459, 1988.

....today are moldable. In fact, in a survey conducted among supercomputer users, 98 of the respondents said that their jobs was moldable [6] SA s goal is to improve job j s turn around time by selecting the request to be sent to the supercomputer. Users provide SA with a set of requests r = r [1] , r [v] that can be used to submit a job j. SA simulates the submission of all requests in r, and then selects the request r [s] that achieves the smallest turn around time in the simulations. The request r [s] is then used to submit job j to the supercomputer. The simulation of ....

....(partition size n, requested time tr, request accuracy a) is known. Let v be number of partition sizes on which j can execute. The moldability model thus uses n, tr, and te to 7 produce the v requests that can be used to submit job j. That is, the moldability model produces the v tuples n = n [1] . n [v] tr = tr [1] tr [v] a = a [1] a [v] that describe v requests for job j. A moldable job is, by definition, a parallel job that can use partitions of various sizes to run. Note, however, that this definition does not mean that a moldable job can run ....

[Article contains additional citation context not shown here]

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1459, 1988.

....draw statistically valid conclusions, as well as to cover the space of parameters we are investigating. Hence, we need a performance metric that summarizes turn around times over multiple experiments. Many researchers have used the mean to combine multiple turn around times into a single metric [Aida 1998] [Cirne 1999] Feitelson 1998b] Krallmann 1999] However, the mean turn around time can be dominated by large jobs [Feitelson 1998b] For example, improving a job s turn around time from 20000 seconds to 18000 seconds (a 10 improvement) reduces the mean by 2000 T, while improving another job s ....

Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1459.

....scheduler is not allowed to choose another job, further in the queue. Because of this restriction, FCFS results in a low maximal processor utilization. In FPFS the scheduler searches further in the queue, from head to tail, and schedules the jobs which fit. It is similar to the backfilling policy [1], but the duration of jobs is not taken into account, so the requirement that the job at the head of the queue should not be delayed is not enforced. In order to avoid starvation (a job is never scheduled) we introduce counters as aging mechanism. Each job counts the number of times it was jumped ....

....phase, and programming difficulties like running out of space for datastructures such as job queues may arise. 0 1 2 3 4 5 6 7 8 9 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0. 8 Utilization Figure 2: The response time for a 32 processor single cluster, with job requests unifo rmly distributed in [1,16], for the FCFS policy. The bars show the 95 confidence intervals) In Table 1 we compare the approximation and the two sets of simulation results for a cluster of 32 processors and for different uniform job size distributions. Overall, the results agree very well, except when the interval of job ....

[Article contains additional citation context not shown here]

K.Aida, H.Kasahara and S.Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science 1459, pages 98-121. Springer-Verlag, 1998.

....order of job size, and then dispatches jobs in the same way as FCFS First Fit. The SJF First Fit dispatches job in the same way as the LJF First Fit except that a job scheduler sorts jobs to the nondecreasing order of job size. Backfilling [2, 3] FCFS fill [11] LSF RTC [2] and FPFS [12] can be classified into this group. 4 Performance Evaluation This section shows simulation results to evaluate performance of job scheduling algorithms under the workload models described in Section 2. The simulation model follows the model described in Section 2. Here, the number of processors ....

....3: Slowdown Ratio (Uniform) enough idle processors are available. Although the priority scheduling also improved performance, they could not prevent blocking as sufficiently as the firstfit scheduling. For further analysis for performance of the first fit scheduling under the Uniform model, see [12]. 4.2 Harmonic Figure 4 and Figure 5 show processor utilization and slowdown ratio respectively under the Harmonic model. There are several differences between results under the Uniform model and those under the Harmonic model. The major difference is observed in performance of the SJF. ....

[Article contains additional citation context not shown here]

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science 1459, pages 98--121, 1998.

No context found.

Kento Aida, Hironori Kasahara, and Seinosuke Narita. Job scheduling scheme for pure space sharing among rigid jobs. In Proceedings of 4th Workshop on Job Scheduling Strategies For Parallel Processing, March 1998.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In D.G. Feitelson and L. Rudolph, editors, 4th Workshop on Job Scheduling Strategies for Parallel Processing, volume 1459 of LNCS, pages 98--121. SpringerVerlag, 1998.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In D.G. Feitelson and L. Rudolph, editors, 4th Workshop on Job Scheduling 1998.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing among Rigid Jobs. In D.G. Feitelson and L. Rudolph, editors, 4th Workshop on Job Scheduling Strategies for Parallel Processing, volume 1459 of LNCS, pages 98--121. Springer-Verlag, 1998.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job scheduling scheme for pure space sharing among rigid jobs. In D. G. Feitelson and L. Rudolph, editors, Job Scheduling Strategies for Parallel Processing, volume 1459 of Lecture Notes in Computer Science, pages 98--121, Berlin Heidelberg New York, 1998. Springer-Verlag.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job Scheduling Scheme for Pure Space Sharing Among Rigid Jobs. In D. Feitelson and L. Rudolph, editors, 4th Workshop on Job Scheduling Strategies for Parallel Processing, volume 1459 of LNCS, pages 98--121. Springer-Verlag, 1998.

No context found.

K. Aida, H. Kasahara, and S. Narita. Job SchedulingScheme for Pure Space Sharing Among Rigid Jobs. In D. Feitelson and L. Rudolph, editors, 4th Workshop on Job Scheduling Strategies for Parallel Processing, volume 1459 of LNCS, pages 98--121. Springer-Verlag, 1998.

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