Kenneth C. Sevcik. Characterizations of Parallelism in Applications and their Use in Scheduling. In Proceedings of the 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... policies consider applications characteristics and propose to execute, for instance, the smallest job first , SJF [Majumdar88] or the largest job first , LJF [Chen88] Other works that also analyze different application characteristics and their use in processor scheduling are [Majumdar91] [Sevcik89] and [Chiang94] However, they are mainly focused on deciding which application must be executed as a function of their execution time. These scheduling policies assume that information about the workload is available a priori, and that the number of processors requested by applications is fixed ....

K. C. Sevcik, "Characterization of Parallelism in Applications and their Use in Scheduling", in Proc. of the ACM SIGMETRICS Conf., pp. 171-180, May 1989.

....the performance as a function of time. It first builds simpler submodels for individual phases, then the totality of submodels describes the performance of the entire execution. Much phase analysis research focuses on how to improve speedup and predict scalability properties. Sevcik in [9] introduced a parallelism profile that characterizes the degree of parallelism (number of busy processors) within an application as function of time. He developed scheduling rules based on four parameters: average, minimum, maximum, and variance parallelism. Several algorithms for automatic phase ....

Sevcik, K. C. Characterization of Parallelism in Applications and Their Use in Scheduling. In Performance Evaluation Review, Vol. 17, No. 1, May 1989

.... 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 parallelism) and or the system (e.g. ....

....28 parallelism) and or the system (e.g. system load) Jobs are typically assumed to be fully moldable in the sense that they can use any number of processors, and the user typically do not provide 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 ....

K. Sevcik. "Characterization of parallelism in applications and their use in scheduling". Proc. of the 1989.

.... policies consider applications characteristics and propose to execute, for instance, the smallest job first , SJF [Majumdar88] or the largest job first , LJF [Chen88] Other works that also analyze different application characteristics and their use in processor scheduling are [Majumdar91] [Sevcik89] and [Chiang94] However, they are mainly focused on deciding which application must be executed as a function of their execution time. These scheduling policies assume that information about the workload is available a priori, and that the number of processors requested by applications is fixed ....

K. C. Sevcik, "Characterization of Parallelism in Applications and their Use in Scheduling ", in Proc. of the ACM SIGMETRICS Conference, pp. 171-180, May 1989.

....by a static analysis of the code and the input file [DI89, B 91, Sar89a] but these techniques often fail outside of toy examples. 6 In a feeble attempt to circumvent the problem, some papers suggest that the users themselves provide some estimates of job parameters [MEB91, Kum88, PD89, Sev89] Even ignoring the fact that this is too far removed from current computing practice, we have to contend with the problem that users will abuse the system by quoting fake values for the parameters. 6 In contrast, estimating the t parameters of the machine (called performance evaluation) is a ....

.... J jobs are not known to be identical in fact, what if nothing is known about them In that case, most papers from academia suggest dynamic space slicing: jobs are space sliced, and as soon as a job terminates the remaining jobs are re allocated to make use of the free PEs [NSS93, R 94, Sev89, ST93, EZL89, SHCV94, LV90, MVZ93, MZ95, TG89, D 95] Unfortunately, in most cases, this result is obtained by facilely neglecting the exorbitant cost of re allocating jobs (see Section 3.7.4) Industry practices gang scheduling: On the other hand, the industry has always leaned towards ....

K.C. Sevcik. Characterization of parallelism in applications and their use in scheduling. In SIGMETRICS Conference on Measurement and Modelling of Computer Systems, pages 171--180, May 1989.

....employed in a number of commercial systems. This is due in part to its low system overhead and its simplicity from both the system and application viewpoints. The static scheduling approach, however, can lead to relatively low system throughputs and resource utilizations under nonuniform workloads [34, 21, 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, ....

K. C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. In Proc. ACM SIGMETRICS Conf., 171--180, 1989.

....employed in a number of commercial systems. This is due in part to its low system overhead and its simplicity from both the system and application viewpoints. The static scheduling approach, however, can lead to relatively low system throughputs and resource utilizations under nonuniform workloads [33, 26, 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, ....

K. C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. In Proceedings of the ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 171--180, May 1989.

....job queues. Internal fragmentation can be a problem with this scheme if jobs are able to utilize a number of processors that does not correspond to an existing partition size. The utility of dividing the available processors into equal sized partitions has been studied with analytic models [Sev89, MEB91, ST93, SST93] compared with adaptive policies [Sev89, MEB91, GST91, RSD 94, ST93] compared with timeslicing policies [SST93] and compared with both adaptive and dynamic policies [NSS93b, NSS93a, NSS97] The general conclusion is that fixed partitioning can perform well if the ....

....this scheme if jobs are able to utilize a number of processors that does not correspond to an existing partition size. The utility of dividing the available processors into equal sized partitions has been studied with analytic models [Sev89, MEB91, ST93, SST93] compared with adaptive policies [Sev89, MEB91, GST91, RSD 94, ST93] compared with timeslicing policies [SST93] and compared with both adaptive and dynamic policies [NSS93b, NSS93a, NSS97] The general conclusion is that fixed partitioning can perform well if the workload is wellunderstood and jobs tend to have low variation in ....

[Article contains additional citation context not shown here]

Kenneth C. Sevcik. Characterizations of Parallelism in Applications and their Use in Scheduling. In Proceedings of the

....so far and the IRIX scheduling policy. Finally, section 6 presents the conclusions of this work. 2 Related Work Many researchers have studied the use of characteristics of the applications calculated at run time to perform processor scheduling. Majumdar et al. [16] Parsons et al. [24] Sevcik [27][28], Chiang et al. [6] and Leutenegger et al. [14] have studied the usefulness of using application characteristics in processor scheduling. They have demonstrated that parallel applications have very different characteristics such as the speedup or 2. In our environment, the multiprogramming level is ....

K.C.Sevcik. "Characterization of Parallelism in Applications and their Use in Scheduling". In Proceedings of the ACM SIGMETRICS Conference, pages 171-180,May 1989

....static precedence structure, as opposed to task graphs with probabilistic structure such as task graph families [25, 83] A workload is an assignment of a duration to each task in the task graph. This dissertation follows the significant body of published work that considers static task durations [29, 34 36, 56, 67]. This is in contrast to the use of stochastic task durations in other work [12, 13, 41, 46, 57, 61, 63, 70] In the future, the main results in this dissertation could be extended to stochastic workloads, continuing on from earlier work in this area [65] In the illustrative task graph shown in ....

K. C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. Proceedings of the

.... 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 are assumed to be fully ....

....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) Moreover, the very lack of a ....

K. Sevcik. Characterization of parallelism in applications and their use in scheduling. Proceedings

....whole area of scheduling with adaptive or dynamic partitioning is based on systems that change the allocation they make as a function of load conditions. Sevcik has proposed adaptive policies that decide on partition sizes based on the load and information about characteristics of the applications [15]. McCann et al. have proposed a dynamic policy that changes the allocation at runtime to reflect changes in the load and requirements [10] Severance et al. propose a scheme that is less dependent on explicit information, in which the system measures the performance of a barrier synchronization to ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....of expensive parallel supercomputers will charge jobs according to their level of parallelism in addition to the total number of CPU cycles consumed. This will prompt users to use the number of PEs that gives the best overall efficiency, rather than that which matches the maximal parallelism [12, 37]. In this paper, we shall use three specific distributions of gang sizes. The first is the uniform distribution. The second is the harmonic distribution, where the probability of size s is proportional 6 to 1=s. Consequently, there will be a larger fraction of small gang sizes. The third is ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....Section 6 concludes our work. 2 2 Related Work As previously mentioned, many researchers have studied the use of application characteristics by processor schedulers of multiprogrammed multiprocessor systems. Majumdar et al. 13] Chiang 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 ....

K. C. Sevcik. Characterizations of Parallelism in Applications and their Use in Scheduling. In Proceedings of the ACM SIGMETRICS Conference, pages 171--180, May 1989.

....of 32 times the optimum for a nonpreemptive 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 ....

K. Sevcik, Characterizations of parallelism in applications and their use in scheduling, in Proceedings of the 1989 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, ACM, New York, 1989, pp. 171--180.

....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 ....

K.C. Sevcik, "Characterization of parallelism in applications and their use in scheduling, " Proc. ACM SIGMETRICS, 1989, pp. 171-180.

....information transparent, performance tuning is extremely difficult. P 3 T at compile time computes a set of performance parameters each of which reflects a different performance aspect. In the following all P 3 T performance parameters are described. 4. 1 Work Distribution It is well known [8, 6, 44, 42, 30, 40, 13, 41, 34, 25] that the work distribution has a strong influence on the cost performance ratio of a parallel system. An uneven work distribution may lead to a significant reduction in a program s performance. Therefore, providing both programmer and parallelizing compiler with a work distribution parameter for ....

K. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171-- 180, 1989.

....various applications. Although multiprogramming allows better service to be provided to the users, it also complicates the processor allocation issues in multiprocessor systems. 2 There are two classes of scheduling algorithms for multiprogrammed parallel systems static and dynamic, 3] 8] 9] [11], 12] The static scheduling algorithms are non preemptive, each application runs to completion without interruption on the set of processors initially allocated for it. The dynamic scheduling algorithms are able to adjust the spatial and the temporal allocations of processing elements to the ....

K. C. Sevcik. Characterization of parallelism in applications and their use in scheduling. Performance Evaluation Review, 17:171--180, May 1989.

....computation phase and communication phase can be identified, which are useful in characterizing workload. For the convenience of analysis, the parallelism profile can be rearranged by accumulating the time spent at each degree of parallelism. The resulting cumulative plot is referred to as shape [Sevcik89] The following figure shows the shape of the parallelism profile shown in Figure 1 (section 3.1) Shape 0 1 2 3 4 5 6 7 8 0 0.2 0.4 0.6 0.8 1 Time ( Processor Figure 2. The shape derived from parallelism profile A shape reflects more intuitively the interested properties (e.g. ....

Sevcik, K. C., "Characterizations of parallelism in applications and their use in scheduling," Proc. ACM SIGMETRICS and Performance'89, pp. 171-180, May 1989.

....[12, 5] there is very little hard evidence from real measurements on parallel machines. The little work that has been done concentrates on modeling single applications, e.g. showing how a specific algorithm leads to changes in the degree of parallelism in different phases of the computation [18, 24, 20]. There is practically no work relating to the mix of different jobs that are found on parallel machines. This makes it hard to study and compare operating system policies for scheduling and processor allocation. Computer Sciences Corporation, NASA Contract NAS 2 12961. This study makes an ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....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 ....

K. Sevcik. Characterizations of Parallelism in Applications and Their Use in Scheduling. In Proceedings of ACM SIGMETRICS Conference, pages 171--180, May 1989.

.... 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 ....

....as far as the kernel can tell. For example, they might be the set of jobs currently residing at the highest non empty priority level in a feedback scheme. While there has been interest in how to use job supplied information on execution characteristics in making scheduling decisions [7, 14, 2], this approach runs contrary to what has proven successful in sequential systems, and relies on the user to be both reliable at estimating job characteristics and honest in relating them to the system. Because we allow jobs to have minimum node requirements, it is not possible for our policies to ....

K.C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. In Proceedings of ACM SIGMETRICS Conference, pages 171--180, May 1989.

....n I i is assigned, indicating the number of components active in the SPG during the respective time interval. Further performance indices can ber derived from a parallelism profile, e.g. the average, minimum and maximum DOP, which are important parameters in mapping and scheduling decisions [13]. From the parallelism profiles, also hypothetical execution times T (n) and speedups S(n) can be derived assuming n available processing elements. A straight forward approach to derive parallelism profiles from TSPGs is to enumerate all possible states of the program (number of active nodes) and ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in schduling", Performance Evaluation Review, Special Issue, 1989 ACM SIGMETRICS, vol. 17, no. 1, pp. 171-- 180, May 1989.

....The key functionality of P 3 T is devoted to compute a set of performance parameters at compile time: ffl Work Distribution The work distribution parameter describes how well the computations of a program are distributed over the set of available processors. As shown by numerous researchers [15, 13, 80, 68, 49, 62, 22, 67, 53, 34], work distribution has a strong influence on the cost performance ratio of a multiprocessor system. An uneven work distribution CHAPTER 3. P 3 T 25 may lead to a significant reduction in a program s performance. Therefore, providing both programmer and compiler with a work distribution ....

K.C. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171--180, 1989.

....for their approach, but they avoid the problem of having the profile change with respect to problem size by analyzing a single problem size over a range of possible workstation networks. Others, like Simon and Wierum [SW96a] require complex microbenchmarks to be run on each system. Ken Sevcik [Sev89] uses a parallelism profile to build schedules for spaceshared machines. His profile is determined experimentally by running the application in the presence of an ample supply of processors. It indicates, as a function of time, the number of processors in use throughout the execution of the ....

Kenneth C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. Performance Evaluation and Review, 17(1):171--180, May 1989.

....n I i is assigned, indicating the number of components active in the SPG during the respective time interval. Further performance indices can ber derived from a parallelism profile, e.g. the average, minimum and maximum DOP, which are important parameters in mapping and schedul4 ing decisions [13]. From the parallelism profiles, also hypothetical execution times T (n) and speedups S(n) can be derived assuming n available processing elements. A straight forward approach to derive parallelism profiles from TSPGs is to enumerate all possible states of the program (number of active nodes) and ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in schduling", Performance Evaluation Review, Special Issue, 1989 ACM SIGMETRICS, vol. 17, no. 1, pp. 171--180, May 1989.

....the example in Figure 5, four different paths are identified. The description of these paths is given in the boxes in Figure 6. 4. Interpretation and Analysis of the Results 4.1. Visualization The results of the second program can be used to derive the potential degree of parallelism (DOP) Sev89] Also the average, the minimum and the maximum parallelism can be computed. In the PV tool different possibilities of visualizing the DOP and the average parallelism are provided. It is also possible to generate non deterministic models, where different paths through a SG are depicted in one ....

K. C. Sevcik. Characterization of parallelism in applications and their use in schduling. Performance Evaluation Review, Special Issue, 1989 ACM SIGMETRICS, 17(1):171--180, May 1989.

....be completly expressed by a data dependency graph or by simplest parameter information such as the average parallelism,or the parallelism profile. Este trabajo ha sido subvencionado por el Ministerio de Educacin bajo los contratos TIC 392 89 y TIC880 92 2 Parallelism profile as defined by [11] is the number of busy processors over the execution time of the program. Average parallelism as defined by Eager,Lazowska and Zahorjan is the average number of busy processors during an execution of the application when an unlimited number of processors are available. We will refer to average ....

Sevcik Kenneth C., "Characterizations of Parallelism in Applications and their use in Scheduling", Performance Evaluation Review Vol.17, May 1989

....information transparent, performance tuning is extremely difficult. P 3 T at compile time computes a set of performance parameters each of which reflects a different performance aspect. In the following all P 3 T performance parameters are described. 4. 1 Work Distribution It is well known [7, 5, 37, 35, 27, 33, 11, 34, 30, 23] that the work distribution has a strong influence on the cost performance ratio of a parallel system. An uneven work distribution may lead to a significant reduction in a program s performance. Therefore, providing both programmer and parallelizing compiler with a work distribution parameter for ....

K. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171-- 180, 1989.

....model. This result for non preemptive 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 ....

K. Sevcik, Characterizations of Parallelism in Applications and their use in Scheduling, Proceedings of the 1989 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, May, 1989, pp. 171--180.

.... comparing static scheduling algorithms a number of researchers have considered characteristics related to the efficiency of a job s execution such as a job s average, minimum, maximum and variation in parallelism, the knee of the execution time efficiency profile, and the processor working set [8, 31, 11, 21]. Eager, Zahorjan, and Lazowska [8] suggest that the number of processors to allocate to each job in a multiprogrammed environment could be made based on the point at which the ratio of efficiency to execution time, E(p i ) T (p i ) is maximized (p i is the number of processors allocated to job J ....

.... assert that policies using execution rate characteristics do not improve performance (compared with policies that use none) especially when the coefficient of variation of job service demand is greater than one (because jobs with long execution times are allocated too many processors) Sevcik [31] finds that static policies that consider additional information about an application s parallelism, such as the minimum, maximum, and variation in parallelism, improve mean response time over methods that only consider average parallelism. Sevcik [32] also identifies a number of useful ....

K. C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. In Proceedings of the 1989 ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 171--180, 1989.

....Section 6 concludes our work. 2 Related Work As previously mentioned,many researchers have studied the use of application characteristics by processor schedulers of multiprogrammed multiprocessor systems. Majumdar et al. 13] Chiang 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 ....

K. C. Sevcik. Characterizations of Parallelism in Applications and their Use in Scheduling. In Proceedings of the ACM SIGMETRICS Conference, pages 171--180, May 1989.

....this with the load information to derive the optimal partition size, that would lead to the best throughput. A simpler scheme uses the program shape, which is the cumulative distribution of the parallelism profile (i.e. what fraction of the time does the program utilize each number of PEs) [520, 341]. This allows the minimal, average, and maximal parallelism to be found. In highly 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 ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....this with the load information to derive the optimal partition size, that would lead to the best throughput. A simpler scheme uses the program shape, which is the cumulative distribution of the parallelism profile (i.e. what fraction of the time does the program utilize each number of PEs) [314, 207]. This allows the minimal, average, and maximal parallelism to be found. In highly 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 ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....database system, NOWs gives more flexibility in terms of processor allocation and load balancing. As a result, query processing on networks of workstations can exploit various processor allocation and load sharing techniques proposed for multiprocessor systems and distributed systems, respectively [7, 8, 10, 12, 16, 17, 19]. Sections 4 and 5 discuss various strategies for load balancing in NOW based parallel query processing. 3.3.2 Job Migration In a NOW based system, the workstations are usually owned by different users. A user may claim his her workstation from the pool of shared workstations for reasons such ....

K.C. Sevcik, "Characterizations of Parallelism in Applications and Their Use in Scheduling," Proc. ACM SIGMETRICS Conference, pp.171-180, May 1989.

....whole are of scheduling with adaptive or dynamic partitioning is based on systems that change the allocation they make as a function of load conditions. Sevcik has proposed adaptive policies that decide on partition sizes based on the load and information about characteristics of the applications [15]. McCann et al. have proposed a dynamic policy that changes the allocation at runtime to reflect changes in the load and requirements [10] Severance et al. propose a scheme that is less dependent on explicit information, in which the system measures the performance of a barrier synchronization to ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

.... Changing A Job s Processor 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 ....

....time. Similar schemes in which co scheduling is triggered by communication events were described by Sobalvarro and Weihl [83] and by Dusseau, Arpaci, and Culler [15] Taking system load and minimum and maximum parallelism of each job into account as well, still higher throughputs can be sustained [77]. Chiang et al. 8] show that use of knowledge of some job characteristics plus permission to use a single preemption per job allows run to completion policies to approach ideal (i.e. no overhead) EQUI, and Anastasiadis et al. 3] show that, by setting the processor allocation of moldable jobs ....

K. C. Sevcik, "Characterization of parallelism in applications and their use in scheduling ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 171--180, May 1989.

....of the one (16) that was used by Rosti et al. 18] 5. 3 The Arrival Process The offered load of a multiprocessor system with P servers is defined as follows : Load = E(T (1) P Theta Mean Interarrival Time (10) 16 where E(T (1) is the mean total execution time of the jobs on one processor [21]. In the case of the jobs that have been used in our study, we have: E(T (1) E(W ) E(fi) E(ff) We already know that E(W ) 13:76. By using the theorem of total expectation across the three different values of p max , we find E(fi) 0:30 and E(ff) 8 ....

Kenneth C. Sevcik. Characterizations of Parallelism in Applications and Their Use in Scheduling. In ACM SIGMETRICS Conf. Measurement and Modeling of Computer Systems, pages 171--180, May 1989.

....set of disciplines that have been proposed and evaluated in the literature. Disciplines presented in this paper are italicized and have the prefix LSF ; for the adaptive ones, a regular and a SUBSET version are provided. Rigid Adaptive Work Speedup Mem. RTC RTC [ZM90] A ,A mM [Sev89] yes min max no PPJ [RSD 94] ASP [ST93] no pws no NQS PWS [GST91] no no no LSF Equal,IP [RSD 94] no no no LoadLeveler SDF [CMV94] yes no no EASY [Lif95] AVG,AdaptAVG [CMV94] no avg no LSF RTC LSF RTC AD(SUBSET) either either either Preemption simple Cosched (matrix) Ous82] ....

Kenneth C. Sevcik. Characterizations of parallelism in applications and their use in scheduling. In Proceedings of the

No context found.

Kenneth C. Sevcik. Characterizations of Parallelism in Applications and their Use in Scheduling. In Proceedings of the 1989.

No context found.

Sevcik, K.C.: Characterizations of Parallelism in Applications and their use in Scheduling. Performance Evaluation Review 17 (1989) 171--180

No context found.

K. C. Sevcik, "Characterizations of Parallelism in Applications and their use in Scheduling," Performance Evaluation Review, vol. 17, no. 1, pp. 171--180, May 1989.

No context found.

Sevcik, K. C. 1989. Characterizations of parallelism in applications and their use in scheduling. In Proc. of ACM Conf. on Measurement and Modeling of Comp. Sys., 171--180.

No context found.

K. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171-- 180, 1989.

No context found.

K.C. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171--180, 1989.

No context found.

K. Sevcik. Characterization of parallelism in applications and their use in scheduling . Performance Evaluation Review, 17(1):171-- 180, 1989.

No context found.

K. C. Sevcik. Characterization of parallelism in applications and their use in schduling. Performance Evaluation Review, Special Issue, 1989 ACM SIGMETRICS, 17(- May 1989.

No context found.

K.C. Sevcik, Characterization of parallelism in applications and their use in schduling, Performance Evaluation Rev., Special Issue, 1989.

No context found.

K. C. Sevcik, Characterizations of Parallelism in Applications and Their Use in Scheduling. Proc.

No context found.

K.C. Sevcik, "Characterizations of Parallelism in Applications and Their Use in Scheduling", ACM Performance Evaluation Review, Vol.17, May 1989.

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