A. B. Downey. Using queue time predictions for processor allocation. 3rd Workshop on Job Scheduling Strategies for Parallel Processing which took place in conjunction with IPPS., 1997.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... allocation of processing intervals, Chiang et al. 9] showed that static processor allocations (for which jobs need only be moldable) led to performance nearly as good as that obtained by dynamic processor allocation (which requires that jobs be malleable) Foregoing Optimal Utilization Downey [12] studies the problem of scheduling in an environment where moldable jobs are activated from an FCFS queue, and run to completion. He suggests how to use predictions of the expected queuing time for awaiting the availability of different numbers of processors in order to decide when a particular ....

....with. Second, even with very many queues, the resolution in which requirements are expressed is necessarily very coarse, and user estimates are notoriously inaccurate anyway. Recent more theoretical work shows how data can be acquired automatically by the system, rather than relying on the users [59,31,12]. At the same time that theoretical work is focusing, at least to some degree, on practical concerns, practice in the field seems to be rather oblivious of this development. One reason is that the larger and more advanced installations have been developing rather elaborate scheduling facilities, ....

A. B. Downey, "Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Springer Verlag, 1997. Lecture Notes in Computer Science (this volume).

....of arrival ia, partition size n, requested time tr, and execution time te. For the jobs that are cancelled by the user, we also want to know their instant of cancellation ic ia. 3.3.1. Instant of Arrival The pattern of job submission is affected by the work cycles of the supercomputer s users [32] [39] 44] For example, typically more jobs are submitted during the day than during the night, as seen in Figure 10 (which indicates how many jobs arrived by hour of the day for the reference workloads) For a workload model to better capture the dynamics of the system, such behavior must be ....

....26 32 1024 32768 1048576 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 CTC Fitted Distribution 32 1024 32768 1048576 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 SDSC Fitted Distribution Figure 13 Cancellation Lag CDF 3.3.3. Partition Size As first noticed by Downey [32] [33] the uniform log distribution also provides a good fit for the partition sizes in a supercomputer workload log. This was the case for our four reference workloads. Table 5 shows the parameters obtained by fitting the observed partition sizes to a uniform log distribution via the method of ....

[Article contains additional citation context not shown here]

Allen Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1291, Dror Feitelson and Larry Rudolph (eds.), 1997. http://www.sdsc.edu/~downey/predalloc/

....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 information about the characteristics of the job (e.g. sequential fraction, average parallelism, and maximum 28 ....

....request times. Strategies that use knowledge about the job have been proposed [3] 8] 12] 25] 28] 30] Adaptation to the system load has also been investigated before [3] 19] 23] Downey has studied whether the job ahead of a FIFO queue should delay its start up to use more processors [10]. Non work conserving strategies were also evaluated by Rosti et al. [29] The results of these efforts indicate that performance improves when processor allocation takes into account job characteristics and system state. Despite such evidence, schedulers currently in use [17] 20] 22] 27] leave ....

A. Downey. "Using Queue Time Predictions for Processor Allocation". Job Scheduling Strategies for Parallel Processing, Springer-Verlag LNCS vol. 1291, 1997.

.... again, sometimes up to hundreds of executions in a row [8, 3] It stands to reason that such repeated executions of the same application would have highly correlated runtimes, and indeed several studies have shown that it is possible to derive crude estimates of runtimes using such information [10, 2, 24]. However, these studies were done in a context that does not penalize underestimation, as is the case with back lling (where jobs that overrun their estimated time are killed) In this context, an estimation method that tends to overestimate is preferred, even if it is less accurate in absolute ....

A. B. Downey, \Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35-57, Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

.... sometimes up to hundreds of executions in a row [8] 3] It stands to reason that such repeated executions of the same application would have highly correlated runtimes and, indeed, several studies have shown that it is possible to derive crude estimates of runtimes using such informa tion [10] [2], 24] However, these studies were done in a context that does not penalize underestimation, as is the case with backfilling (where jobs that overrun their estimated time are killed) In this context, an estimation method that tends to overestimate is preferred, even if it is less accurate in ....

A.B. Downey, "Using Queue Time Predictions for Processor Allocation," Job Scheduling Strategies for Parallel Processing, D.G. Feitelson and L. Rudolph, eds., pp. 35-57, Springer-Verlag, 1997.

....delay and 6 dim jobs in a 6 cube hypercube do not either, since a 6 dim job demands all processors of the hypercube computer. Predicting the queue delay involves applying probabilistic model to the busy processors. There have been several attempts to predict queue delay analytically [10] 11][12] and the prediction plays in important role in making decisions on processor allocation. The processor based prediction counts the number of busy processors and analyzes the queue delay based on the number[10] 12] use the number of processors to predict queue time in the em conditional life ....

....There have been several attempts to predict queue delay analytically [10] 11] 12] and the prediction plays in important role in making decisions on processor allocation. The processor based prediction counts the number of busy processors and analyzes the queue delay based on the number[10] [12] use the number of processors to predict queue time in the em conditional life model. One of the disadvantages of the processor based prediction can be explained with Figure 2. When a dimension 5 job, J4, arrives on a 6 cube hypercube parallel computer, it can be allocated to the space S1 or S2 ....

[Article contains additional citation context not shown here]

A.Downey, "Using Queue Time Prediction for Processor Allocation", Job Scheduling Strategies for Parallel Processing, Lecture Notes in Computer Science, Vol.1291, pp.35-57, 1997.

....data. They find that more accurate estimates of job execution time lead to more accurate estimates of wait time. The authors do observe an improvement in average job wait time, for a particular Argonne National Laboratory workload, when using their predictors instead of previously published work [1, 9]. 4 Gang scheduling In the previous sections we only considered space sharing scheduling strategies. An extra degree of flexibility in scheduling parallel jobs is to share the machine resources not only spatially but also temporally by partitioning the time axis into multiple time slices [2, 4, ....

A. B. Downey. Using Queue Time Predictions for Processor Allocation. In IPPS'97 Workshop on Job Scheduling Strategies for Parallel Processing, volume 1291 of Lecture Notes in Computer Science, pages 35--57. Springer-Verlag, April 1997.

....detail in Chapter 4. The other parameters are discussed below. Chapter 2. Background 35 2.2.1 Job Inter arrival Time The most popular choice for modeling job arrivals has been a Poisson process. Numerous studies have used this distribution [MEB88, LV90, ZM90, GST91, CMV94, PS95, CV96, Fei96, Dow97] although some have experimented with the hyper exponential distribution [RSD 98, CV96, IPS96] An exponential distribution is known to have a CV of 1. However, studies of workload traces from parallel computing sites have tended to find inter arrival CV s considerably larger than 1. In the ....

....Many studies [FN95, SSG95, Hot96, SGS96, WMKS96] have also noted the large numbers of small, short jobs often over half the total number of jobs. As well, they have all found that these small jobs consume a small fraction of the total resources, e.g. node seconds or CPU cycles. Downey [Dow97] proposes a model where the sequential lifetime L (runtime of a job on a single processor) is used to calculate the speedup of a job, and thus the runtime on n processors. Since L is generally not known, T (the wall clock time multiplied by Chapter 2. Background 37 the cluster size) is used as an ....

[Article contains additional citation context not shown here]

Allen B. Downey. Using queue time predictions for processor allocation. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 35--57. Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

....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 (e.g. sequential fraction, average parallelism, and maximum parallelism) and or the system (e.g. system load) Jobs ....

....characteristics and system state. Despite such evidence, scheduling solutions currently in use [20] 23] 25] 28] leave processor allocation to the user by accepting only rigid requests. We believe that this is due to the difficulty in proving that proposed processor allocation solutions [3] 5] [12] [14] 18] 29] 30] 31] 32] will work in practice. Such difficulty happens because existing work are more theoretical in nature and thus, in order to keep things tractable, make use of strong assumptions (e.g. Poisson job arrival, full moldability, and accurate information about jobs) ....

A. Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Feitelson and Rudolph (Eds.), Springer-Verlag Lecture Notes in Computer Science, vol. 1297, 1997. http://www.sdsc.edu/~downey/predalloc/

.... In current practice, supercomputer schedulers accept rigid requests [18] 21] 22] 25] and thus much of the research available in the literature assume jobs to be rigid [1] 2] 14] 17] 27] 29] However, there is evidence that performance can be improved by allowing jobs to be moldable [11], as in this paper, or even malleable [26] In [11] the supercomputer scheduler is assumed to be FIFO and moldability is exploited by the first job in the queue to decide whether it is worth to wait for more processors to be available, reducing therefore the execution time (but increasing the ....

.... accept rigid requests [18] 21] 22] 25] and thus much of the research available in the literature assume jobs to be rigid [1] 2] 14] 17] 27] 29] However, there is evidence that performance can be improved by allowing jobs to be moldable [11] as in this paper, or even malleable [26] In [11], the supercomputer scheduler is assumed to be FIFO and moldability is exploited by the first job in the queue to decide whether it is worth to wait for more processors to be available, reducing therefore the execution time (but increasing the wait time) In [26] jobs are assumed to be malleable, ....

Allen Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1291, Dror Feitelson and Larry Rudolph (eds.), 1997. http://www.sdsc.edu/~downey/predalloc/

.... current practice, supercomputer schedulers accept rigid requests [17] 20] 22] 25] and thus much of the research available in the literature assume jobs to be rigid [1] 2] 14] 16] 21] 28] 30] However, there is evidence that performance can be improved by allowing jobs to be moldable [4] [11], as in this paper, or even malleable [26] Despite such evidence, in practice, deploying scheduling solutions that allow jobs to be non rigid has been slow. We believe that this is due to the difficulty in proving that a given scheduling solution for moldable or malleable jobs will work in ....

A. Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1291, Dror Feitelson and Larry Rudolph (eds.), 1997. http://www.sdsc.edu/~downey/predalloc/

....we have to determine performanceefficient values of R and S for each available supercomputer N . Our goal is to select and S in a way that minimizes GTOMO s turnaround time. Note that difficulty in predicting supercomputer queue wait times make it difficult to find an optimal R and S [8, 20, 14]. We avoid the queue time prediction problem by using supercomputer nodes that are immediately available. Therefore, we minimize the turnaround time of GTOMO by scheduling its execution at once on workstations and any immediately available supercomputer nodes. We assume that the supercomputer ....

A. Downey. Using queue time predictions for processor allocation. In 3rd Workshop on Job Scheduling Strategies for Parallel Processing, in conjunction with IPPS'97, 1997.

....Downey s parameters A and s (as we shall see in Section 4) and hence does not need to be modeled directly. 3.1. Partition Size Constraints As Table 1 shows, the distribution of partition sizes are dominated by power of 2 values in many workload logs. This phenomenon was also observed by others [7] [9] 13] 20] There has been some controversy as to whether to incorporate the dominance of power of 2 partitions into a workload model. Some researchers have accounted for the high incidence of power of 2 jobs and modeled the partition size accordingly [13] Others, however, believe that this ....

.... the partition size accordingly [13] Others, however, believe that this phenomenon is mainly an artifact of behavioral inertia (the first parallel supercomputers required power of 2 partition sizes) and the design of some submission interfaces (which suggest the submission of power of 2 jobs) [7]. Based solely on workload logs, it is impossible to determine whether the prevalence of power of 2 jobs is due to the nature of the parallel jobs, or is an artifact of behavioral inertia and interface design. In the survey, we inquired about the constraints jobs have regarding partition size ....

[Article contains additional citation context not shown here]

Allen Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1291, Dror Feitelson and Larry Rudolph (eds.), 1997. http://www.sdsc.edu/~downey/predalloc/

.... again, sometimes up to hundreds of executions in a row [8, 3] It stands to reason that such repeated executions of the same application would have highly correlated runtimes, and indeed several studies have shown that it is possible to derive crude estimates of runtimes using such information [10, 2, 24]. However, these studies were done in a context that does not penalize underestimation, as is the case with back lling (where jobs that overrun their estimated time are killed) In 26 OK jobs w info OK jobs no info killed jobs 0 10 20 30 40 50 60 70 80 90 100 110 120 130 percent of requested ....

A. B. Downey, \Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35-57, Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

....that includes cost and how SA pursues such a goal are also intriguing research questions. 6. Related Work There has been great interest in supercomputer scheduling in recent years. Some of the research in this area allow for the scheduler to choose the number of processors allocated to a job [Downey 1997b] or even to change this number during the execution of the job [Chiang 1996] Nguyen 1996] Such schedulers therefore try to improve the performance of the system in the same way SA aims to improve the performance of the job. The main distinction is exactly that these efforts take the system wide ....

Allen B. Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, SpringerVerlag, Lect. Notes Comput. Sci. vol. 1162, 1997. KWWS###ZZZ#VGVF#HGX#aGRZQH\#SUHGDOORF# 19

....the first, we have to determine performanceefficient values of n and t for each available supercomputer s. Our goal is to select n and t in a way that minimizes GTOMO s turnaround time. Note that difficulty in predicting supercomputer queue wait times make it difficult to find an optimal n and t [8, 20, 14]. We avoid the queue time prediction problem by using supercomputer nodes that are immediately available. Therefore, we minimize the turnaround time of GTOMO by scheduling its execution at once on workstations and any immediately available supercomputer nodes. We assume that the supercomputer ....

A. Downey. Using queue time predictions for processor allocation. In 3rd Workshop on Job Scheduling Strategies for Parallel Processing, in conjunction with IPPS'97, 1997.

....[50] However, today there is no such tool for space sliced parallel supercomputers. Since jobs run on a dedicated set of nodes in these machines, the information application schedulers don t have today regards the queue waiting time. In principle, work on supercomputer queue time prediction [11,46,25] could be used to provide this information. However, the results obtained for queue time predictions are still relatively innaccurate, making them inadequate for many metacomputing applications, notably those that perform co allocation (i.e. that spread across multiple resource schedulers) This ....

A. B. Downey, "Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35--57, Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

....importance of the jobs. This techniques needs both user selected and administrative priorities which are currently not available to Winner s scheduler. Another possible improvement could so far not be implemented in Winner s batch scheduler due to missing information. So called malleable jobs [9] are parallel jobs that can run with different numbers of processors from a given interval, and possibly use additional CPUs in the middle of the run when they become available. The presence of such jobs would give a large degree of freedom to Winner s backfilling scheduler but would also ....

A.B. Downey. Using Queue Time Predictions for Processor Allocation. In Proc. International Parallel Processing Symposium (IPPS'97) Workshop on Job Scheduling Strategies for Parallel Processing, 1997.

....more robust than the others, meaning that on average it performs better for widely different load conditions and for different workload characteristics. A more sophisticated approach is to base the decision on predictions of the queueing time, based on estimates of how long current jobs will run [164]. This allows the system to weigh the two alternatives: either allocate a small number of PEs immediately, leading to a longer runtime, or wait for more PEs to become available, hopefully leading to shorter execution and a faster overall response time. Setting the partition size in cooperation ....

....are faced with uncertainty regarding when they will be able to run their programs. Reducing the uncertainly is just as important as reducing the response time itself. Methods to do so include placing limits on running jobs, as is done in NQS, and trying to estimate the queueing time of new jobs [368, 225, 164]. Some partitioned systems provide batch scheduling as the only support to ease these difficulties. Others do not provide even this limited support, and implicitly expect users to coordinate the use of the machine verbally among themselves [151, 303] This is not always easy, as different users ....

[Article contains additional citation context not shown here]

A. B. Downey, "Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35--57, Springer Verlag, 1997. Lecture Notes in Computer Science Vol. 1291.

....shorter, the transition is to a state with larger partitions. This reduces the danger of leaving idle PEs while some applications are still active. A more sophisticated approach is to base the decision on predictions of the queueing time, based on estimates of how long current jobs will run [93]. This allows the system to weigh the two alternatives: either allocate a small number of PEs immediately, leading to a longer runtime, or wait for more PEs to become available, hopefully leading to shorter execution and a faster overall response time. Setting the partition size in cooperation ....

....the workload. So far systems simply had to make do without such information, or else users had to supply it in the form of bounds on resource use. There are interesting research opportunities in the area of automatic classification of parallel jobs, and assessment of their resource requirements [140, 93]. This should probably involve some cooperation between the compiler and the runtime system. In addition, there is a dire need for information about statistical properties of parallel workloads in general, e.g. the distribution of partition sizes that are requested in a general purpose ....

A. B. Downey, "Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35--57, Springer Verlag, 1997. Lecture Notes in Computer Science Vol. 1291.

....the first, we have to determine performance efficient values of n and t for each available supercomputer s. 3 Our goal is to select n and t in a way that minimizes GTOMO s turn around time. Note that difficulty in predicting supercomputer queue time makes it difficult to find an optimal n and t [Dow97, STF99, Gib97] We avoid the queue time prediction problem by using supercomputer nodes that are immediately available. We assume that the supercomputer scheduler can provide us with the maximum values of n and t for which execution can begin immediately. Our implementation uses the showbf ....

Allen Downey. Using queue time predictions for processor allocation. In 3rd Workshop on Job Scheduling Strategies for Parallel Processing, in conjunction with IPPS'97, 1997.

....elapsed between submission and completion: the sum of queue wait time and execution time in the batch case) However, difficulty in predicting supercomputer queue time makes it difficult to find the optimal n and t. For some efforts in predicting supercomputer execution and queue wait time, see [Dow97, STF99, Gib97] We avoid this problem by simply using the nodes that are currently idle. We assume that the supercomputer scheduler provides us with the maximum values of n and t for which execution will begin immediately. Therefore, the AppLeS application scheduler can make a well founded ....

Allen Downey. Using queue time predictions for processor allocation. In 3rd Workshop on Job Scheduling Strategies for Parallel Processing, in conjunction with IPPS'97, 1997.

....resources (e.g. 61] Today there is no such tool for space sliced parallel supercomputers. Since jobs run on a dedicated set of nodes in these machines, the information metaschedulers can expect to obtain regards the queue waiting time. In principle, work on supercomputer queue time prediction [15,57,31] could be used to provide this information. However, the results obtained for queue time predictions are still relatively inaccurate, making them inadequate for many metacomputing applications, notably those that perform co allocation (i.e. that spread across multiple machine schedulers) This ....

A. B. Downey, "Using queue time predictions for processor allocation". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35--57, Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

....nearly as good as that obtained by dynamic processor allocation (which requires that jobs be malleable) 10] 2.2. 5 Non work conserving scheduling Downey has studied the problem of scheduling in an environment where moldable jobs are activated from an FCFS queue, and run to completion [13]. He suggests how to use predictions of the expected queuing time for awaiting the availability of different numbers of processors in order to decide when a particular job should be activated. The tradeoff is between starting a job sooner with fewer processors and delaying its start (causing ....

....with. Second, even with very many queues, the resolution in which requirements are expressed is necessarily very coarse, and user estimates are notoriously inaccurate anyway. Recent more theoretical work shows how data can be acquired automatically by the system, rather than relying on the users [60, 32, 13]. At the same time that theoretical work is focusing, at least to some degree, on practical concerns, practice in the field seems to be rather oblivious of this development. One reason is that the larger and more advanced installations have been developing rather elaborate scheduling facilities, ....

A. B. Downey, "Using queue time predictions for processor allocation". In IPPS'97 Workshop Job Scheduling Strategies for Parallel Processing.

....consider the problem of contention with other jobs in the system, but they do not have the problem of predicting the time until a cluster becomes available. Like us, they raise the question of how local optimization by greedy users affects overall system performance. We address this question in [5]; we allow each job to choose the cluster size that minimizes its expected turnaround time, and find that this application centric strategy leads to global performance better than that of many proposed system centric strategies. 3 Conditional lifetime model (CLM) During 1995, 24906 jobs were ....

A. B. Downey. Using queue time predictions for processor allocation. Technical Report CSD-97-929, University of California at Berkeley, 1997. To appear at the Job Scheduling Strategies for Parallel Programming Workshop at IPPS '97.

....CV of the distribution of areas, we would rather say 3.6 (the CV of the fitted model) than, Something less than 29.9, depending on how many of the data turn out to be bogus. In prior work, Downey used these techniques to summarize the log uniform distributions that are common on batch systems [8], and Harchol Balter and Downey used a similar technique to summarize the Pareto distributions that are commonon sequential interactive systems [17] 3.4 Goodness of fit The above description is based on the eyeball method: we transformed the CDF into log space, observed that the result looked ....

A. B. Downey, "Using queue time predictions for processor allocation ". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 35--57, Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

....on various cluster sizes (subject to simplifying assumptions about the program s behavior) When oe = 0, our model matches the upper bound; as oe approaches infinity, our model approaches the lower bound asymptotically. This model might be used differently for different applications. In [5] and [6] we use it to generate the stochastic workload we use to evaluate allocation strategies for malleable 1 jobs. For that application, we choose the parameters A and oe from distributions and use them to generate speedup curves. In this paper, we work the other way around we use observed ....

....models based on Dowdy s speedup model have tended to overestimate the parallelism available in codes executing in supercomputing environments. With our model, we have been able to use observations of the workload at the San Diego Supercomputer Center to infer the parameters of real workloads [5] [6]. Sevcik [16] and Ghosal et al. 8] have proposed alternative models based on more detailed program information. These models have many free parameters, and therefore provide no way to infer program characteristics from observed behavior. Furthermore, it would be difficult to specify the range of ....

Allen B. Downey. Using queue time predictions for processor allocation. Technical Report CSD-97-929, University of California at Berkeley, 1997.

No context found.

A. B. Downey. Using queue time predictions for processor allocation. 3rd Workshop on Job Scheduling Strategies for Parallel Processing which took place in conjunction with IPPS., 1997.

No context found.

Allen Downey. Using Queue Time Predictions for Processor Allocation. In Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science Vol. 1291, Dror Feitelson and Larry Rudolph (eds.), 1997. KWWS###ZZZ#VGVF#HGX#aGRZQH\#SUHGDOORF#

No context found.

A. Downey. Using queue time predictions for processor allocation. In 3rd Workshop on Job Scheduling Strategies for Parallel Processing, in conjunction with IPPS'97, 1997.

No context found.

Downey, A, B., (1997), "Using Queue Time Predictions for Processor Allocation", Job Scheduling Strategies for Parallel Processing (JSSPP'97), Geneva, Switzerland, April 5, Lecture Notes in Computer Science (LNCS) 1291, Springer Verlag, pp. 35-57.

No context found.

Allen B. Downey. Using queue time predictions for processor allocation. In Dror G. Feitelson and Larry Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 35--57. Springer Verlag, 1997. Lect. Notes Comput. Sci. vol. 1291.

No context found.

Allen B. Downey. Using Queue Time Predictions for Processor Allocation. In D. G. Feitelson and L. Rudolph, editors, Job Scheduling Strategies for Parallel Processing, pages 35-- 57. Springer-Verlag, 1997.

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