K. Dussa, K. Carlson, L. Dowdy, and K-H. Park, "Dynamic partitioning in a transputer environment". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 203--213, May 1990.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Several studies have revealed that EQUI does very well, even when some moderate charge for the overhead of frequent preemptions is made [86,48] Squillante [84] provides an analysis of the performance of dynamic partitioning. Deng et al. show that EQUI is optimally competitive [11] Dussa et al. [14] compares space slicing against no partitioning, and finds that space partitioning pays off. Knowledge Based Scheduling Majumdar, Eager and Bunt showed that, under high variability service time distributions, round robin (RR) was far better than FCFS, but that policies based on knowledge of the ....

K. Dussa, K. Carlson, L. Dowdy, and K-H. Park, "Dynamic partitioning in a transputer environment". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 203--213, May 1990.

....processors during execution. In distributed systems, the applications must dynamically redistribute their data structures. This involves considerable overhead, which can outweigh the benefits of the processor reallocation itself, as has been demonstrated both analytically and experimentally [6, 17]. Sophisticated run time systems are required to do dynamic data redistribution automatically, and keep it from being an additional burden on the application writer [3, 8, 16] Recently, it has been shown by Feitelson and Nitzberg [9] that application characteristics, such as execution time on a ....

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park. Dynamic Partitioning in a Transputer Environment. In ACM SIGMETRICS Conf. Measurement and Modeling of Computer Systems, pages 203--213, May 1990.

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

.... environments (e.g. non uniform access, shared memory and distributed memory systems) however, the overheads of a dynamic partitioning policy can be significant due to factors such as data job migration, processor preemption coordination and, in some cases, reconfiguration of the application [4, 21, 22, 30]. Even with continuing reductions in the latency of interprocessor communication [41, 39] there are other factors that can cause the cost of repartitioning to be significant for important classes of scientific engineering applications (e.g. the need to reconfigure the application) 19, 20, 18] ....

[Article contains additional citation context not shown here]

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park. Dynamic partitioning in transputer environments. In Proc. ACM SIGMETRICS Conf., 203--213, 1990.

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

.... strategies in such environments [37, 21, 38, 17, 23] In more distributed parallel environments, however, the overheads of a dynamic partitioning policy can be significant due to factors such as data job migration, node preemption coordination and, in some cases, reconfiguration of the application [9, 26, 27, 31]. There are several fundamental issues that must be considered in order to effectively exploit dynamic partitioning in distributed computing environments. First, the applications must be capable of executing on variable numbers of nodes and must be capable of reconfiguring the number of nodes on ....

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park. Dynamic partitioning in transputer environments. In Proceedings of the ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 203--213, 1990.

....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. Dussa, B.M. Carlson, L.W. Dowdy, K.-H. Park, "Dynamic partitioning in a transputer environment," Proc. ACM SIGMETRICS, 1990, pp. 203-213.

....the needs of both interactive applications belonging to workstation owners and multiple batch applications trying to scavenge idle cycles from the cluster. One of the scheduling policies that has been shown to have good performance in dedicated parallel environments is dynamic space sharing [6, 13, 17]. Under dynamic space sharing, the processors of a parallel system are divided into disjoint partitions that are allocated to individual jobs. However, the number of processors allocated to a partition can be changed dynamically in response to events such as new job arrivals or job departures. In ....

....applications, and presents measurement results for the cost of dynamic reconfiguration. Section 5 describes future work and conclusions. 2 Motivation for Dynamic Spacesharing In the last few years, several studies have examined job scheduling strategies for dedicated parallel environments [6, 13, 11, 7, 17]. Policies that employ dynamic space sharing have been shown to outperform other policies such as static space sharing and gangscheduling with static partitioning for many workloads. In addition to its advantages in dedicated environments, dynamic space sharing has some natural advantages in ....

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park. Dynamic partitioning in transputer environments. In Proc. of the ACM SIGMETRICS Conf., 1990.

....to a dynamically changing number of PEs is possible in other programming models as well, but it requires significant effort on the side of the program developer. For example, jobs written for distributed memory machines can adjust to a changing number of PEs by redistributing their data structures [422, 171, 396]. In some cases, such redistribution is supported by the system (albeit with the intention of redistribution for different phases of the computation, not for changes in the available PEs) 23, 378] As this involves considerable overhead, it is imperative that the frequency of reconfigurations be ....

.... phases of the computation, not for changes in the available PEs) 23, 378] As this involves considerable overhead, it is imperative that the frequency of reconfigurations be kept low in practice: otherwise the price of reconfiguration might even outweigh the benefits of changing the PE allocation [446, 171, 543]. Alternatively, a model where reconfigurations are only allowed at certain points in the application can be used [630, 629, 185, 521] These points are chosen such that repartitioning is significantly easier than at other points in the computation, e.g. at the beginning of a new parallel loop 8 ....

K. Dussa, K. Carlson, L. Dowdy, and K-H. Park, "Dynamic partitioning in a transputer environment". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 203-- 213, May 1990.

....to a dynamically changing number of PEs is possible in other programming models as well, but it requires significant effort on the side of the program developer. For example, jobs written for distributed memory machines can adjust to a changing number of PEs by redistributing their data structures [260, 97, 239]. As this involves considerable overhead, it is imperative that the frequency of reconfigurations be kept low in practice: otherwise the price of reconfiguration might even outweigh the benefits of changing the PE allocation [272, 97, 327] Alternatively, a model where reconfigurations are only ....

.... number of PEs by redistributing their data structures [260, 97, 239] As this involves considerable overhead, it is imperative that the frequency of reconfigurations be kept low in practice: otherwise the price of reconfiguration might even outweigh the benefits of changing the PE allocation [272, 97, 327]. Alternatively, a model where reconfigurations are only allowed at certain points in the application can be used [365, 364, 109, 315] These points are chosen such that repartitioning is significantly easier than at other points in the computation, e.g. at the beginning of a new parallel loop 6 ....

K. Dussa, K. Carlson, L. Dowdy, and K-H. Park, "Dynamic partitioning in a transputer environment". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 203-- 213, May 1990.

.... does very well, even when some moderate charge for the overhead of frequent preemptions is made [87, 49] Squillante has provided an analysis of the performance of dynamic partitioning [85] Dussa et al. compared space slicing against no partitioning, and found that space partitioning pays off [15]. On the other hand, coscheduling is compared to local scheduling and is found to be superior by Dusseau, Arpaci, and Culler [16] 2.2.7 Knowledge based scheduling Majumdar, Eager and Bunt showed that, under high variability service time distributions, round robin (RR) was far better than FCFS, ....

K. Dussa, K. Carlson, L. Dowdy, and K-H. Park, "Dynamic partitioning in a transputer environment ". In SIGMETRICS Conf. Measurement & Modeling of Comput. Syst., pp. 203--213, May 1990.

....wait in first come first served order and run to completion when activated. Research in scheduling for distributed memory architectures, however, has focussed on dynamic partitioning, where processors are dynamically reallocated in response to the number of jobs in the system [MZ94, NSS93, DCDP90] In work by McCann and Zahorjan, applications are assumed to be designed for a virtual machine that matches the characteristics of the physical machine, which in this case is a mesh. As the system load increases, the threads of a job are multiplexed in such a way that the load on each processor ....

....scheme, imbalances in processor allocation are evened out in the long run. A similar approach was also used in an earlier study of a two job workload running on a small transputer based system arranged in a ring, where it was concluded that dynamic partitioning was both feasible and desirable [DCDP90] Others have demonstrated that application redistribution, as required by equipartitioning, is also feasible at moderate cost, and that under high variability in service demand, can offer significantly better performance than a static partitioning scheme [NSS93, Set93] But the high cost of ....

K. Dussa, B. Carlson, L. Dowdy, and K-H. Park. Dynamic partitioning in a transputer environment. In Proceedings of the 1990 ACM SIGMETRICS Conference on Measurement and Modelling of Computer Systems, pages 203--213, 1990.

.... different parallel architectures, task structures, communication issues and load balancing [8, 13] Typically, experimentally observed performance (e.g. speedup or response time) is tabulated as a function of the number of processors employed, a function sometimes known as the execution signature [10], or response time function. In this paper we use such functions to determine the number of processors to be allocated to each of several tasks when the tasks are part of a pipelined computation. This problem is natural, given the growing availability of multitasked parallel architectures, such as ....

....the computation, so that different tasks are concurrently applied to different data sets. Each task is potentially parallelizable; for each t i we let f i (n) be the execution time of t i using n identical processors. f i is called a response time function (also known as an execution signature [10]) We assume that f 0 and f n 1 are dummy tasks that serve respectively to identify the initiation and completion of the computation; correspondingly we take f 0 (n) f n 1 (n) 0 for all n. However, f i (0) 1 for all i = 1; n; these conditions ensure that no processor is ever ....

K. Dussa and B. Carlson and L. Dowdy and K.-H. Park. Dynamic partitioning in a transputer environment. Proceedings of the 1990 ACM SIGMETRICS Conference, 203--213, May 1990.

....the needs of both interactive applications belonging to workstation owners and multiple batch applications trying to scavenge idle cycles from the cluster. One of the scheduling policies that has been shown to have good performance in dedicated parallel environments is dynamic space sharing [7, 19, 23, 31]. Under dynamic space sharing, the processors of a parallel system are divided into disjoint partitions that are allocated to individual jobs. However, the number of processors allocated to a partition can be changed dynamically in response to events such as new job arrivals or job departures. In ....

....results for the cost of dynamic reconfiguration. Section 5 discusses related work, and Section 6 describes future work and conclusions. 2 Motivation for Dynamic Space sharing In the last few years, several studies have examined job scheduling strategies for dedicated parallel environments [7, 19, 16, 10, 23]. Policies that employ dynamic space sharing have been shown to outperform other policies such as static space sharing and gang scheduling with static partitioning for many workloads. The main reasons for this are that (i) dynamic space sharing can react better to changes in system load, and (ii) ....

[Article contains additional citation context not shown here]

K. Dussa, B. Carlson, L. Dowdy, and K-H. Park. Dynamic partitioning in transputer environments. In Proc. of the ACM SIGMETRICS Conf., 1990.

....processor allocation for cube connected message passing machines. Like Feitelson and Rudolph, they assume that arriving jobs declare the number of processors required for execution. The goal of their work is to describe efficient schemes for finding sub cubes to satisfy arriving jobs. Dussa et al. [4] examine the benefits of dynamically repartitioning processors on job arrival and departure, using experiments on a ring connected Transputer based system, as well as a simple analytic model. They conclude that for this hardware and the two job workload considered that dynamic repartitioning can ....

K. Dussa, B. Carlson, L. Dowdy, and K-H. Park. Dynamic partitioning in a transputer environment. In Proceedings of ACM SIGMETRICS Conference, pages 203--213, May 1990.

....limited due to its inability to adjust scheduling decisions in response to subsequent workload changes. These potential problems are alleviated under dynamic partitioning, where the size of the partition allocated to a job can be modified during its execution, at the expense of increased overhead [39, 9, 21, 40, 16, 23, 26, 27, 36, 28]. The relative runtime costs of a dynamic partitioning policy are heavily dependent upon the parallel architecture and application workload under consideration. In uniform access, shared memory (UMA) systems, these overheads tend to be relatively small and thus the benefits of dynamic partitioning ....

.... strategies in UMA environments [39, 21, 40, 16, 23] In more distributed parallel environments, however, the overheads of a dynamic partitioning policy can be significant due to factors such as data job migration, node preemption coordination and, in some cases, reconfiguration of the application [9, 26, 27, 33, 28]. There are several fundamental issues that must be considered in order to e#ectively exploit dynamic partitioning in distributed parallel computing environments. First of all, the applications must be capable of executing on variable numbers of nodes and must be capable of reconfiguring the ....

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park. Dynamic partitioning in transputer environments. In Proceedings of the ACM SIGMETRICS Conference on Measurement and Modeling of Computer Systems, pages 203--213, 1990.

....is only a single level to the allocation structure in pool based scheduling. Additionally, in pool based scheduling the kernel can effectively restrict a job s choice of parallelism by restricting its threads to only one or a few pools. Pool based scheduling may time share partitions. Dussa et al. [15] examine the benefits of dynamically repartitioning processors on job arrival and departure, with experiments on a ring connected Transputer based system, as well as a simple analytic model. They conclude that for this hardware and the two job workload considered that dynamic repartitioning can be ....

K. Dussa, B. Carlson, L. Dowdy, and K-H. Park. Dynamic partitioning in a transputer environment. In Proceedings of ACM SIGMETRICS Conference, pages 203-- 213, May 1990.

....overhead is small we can realize performance benefits by using a dynamic policy. This is typically the case for uniform memory access (UMA) shared memory systems but the overhead is not small for distributed memory systems. However, dynamic policies have been proposed for distributed memorysystems [7, 17]. 2.2 Time Sharing Policies A special type of dynamic policies are time sharing policies that use preemption to rotate processors among the jobs. There are two principal ways in which preemption can be done: coordinated or uncoordinated. In coordinated preemption, also called coscheduling [19] ....

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park, "Dynamic Partitioning in a Transputer Environment", Proceedingsof the ACM SIGMETRICS, May 1990, pp. 203-213.

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

....detailed knowledge of various parameters of the system under consideration is necessary. Also, validation of the simulation models is difficult. Detailed Markovian analysis of complex scheduling policies to verify simulation models is possible only for small systems (e.g. less than 10 processors) [SRSDS94, DCDP90, MEB88]. For larger systems, simplifying assumptions have to be made, resulting in a loss of accuracy. Experimental studies on dynamic processor partitioning policies have been done mainly for shared memory architectures [GTU91, TG89] In this paper, experimental analysis of dynamic processor ....

K. Dussa, B.M. Carlson, L.W. Dowdy, K.-H. Park, "Dynamic partitioning in a transputer environment," Proc. ACM SIGMETRICS, 1990, pp. 203-213.

....allocated to an application is allowed to change during its execution. The number of processors assigned to an application changes depending on the system load [TG89, MZ94] or on changes in the applications parallelism [ZM90] Preemptive space sharing policies are also known as dynamic policies [TG89, DCDP90, ZM90]. Hybrid partitioning schemes are also preemptive, where the multiprocessor is divided into possibly non disjoint partitions. In each partition, tasks of different applications execute in time sharing mode [ZB91, SST93] The focus of this paper is on non preemptive policies as they represent a ....

K. Dussa, B. M. Carlson, L. W. Dowdy, and K.-H. Park, "Dynamic partitioning in a transputer environment," Proc. ACM SIGMETRICS, 1990, pp. 203-213.

....by allocating less than the entire complement of processors to a job. The improvement results from allowing several parallel jobs to share the multiprocessor simultaneously. The processor allocation problem is to determine the appropriate number of processors to allocate to each parallel program [1 7]. Three types of processor allocation strategies exist. A static allocation strategy partitions the processors at system generation time. When a job arrives, it waits for an available partition and then executes until completion using all of the processors in its assigned partition. A dynamic ....

K. Dussa, B. Carlson, L. W. Dowdy, and K. H. Park, "Dynamic partitioning in a transputer environment," in Proc., ACM Sigmetrics Conference on Measurement and Modeling of Computer Systems, pp. 203--213, May 1990.

....basis [6] Space sharing involves partitioning the processors among the parallel jobs. Partitioning schemes can be either fixed or variable. In the first case, the processor partitioning is divided into fixedsized partitions. The partition sizes are set at system generation time and do not change [12]. This scheme is appropriate for fixed workload environments. With variable partitioning, the size of the partition allocated to each job varies, according to the job s characteristics and scheduling policy. Variable partitioning is implemented by the allocation phase of scheduling disciplines, ....

Dussa, K., Carlson, B., Dowdy, L. W., and Park, K. Dynamic partitioning in a transputer environment. Proc. ACM Sigmetrics Conference, 1990.

....application parallelism, diminishing return from assigning additional processors, and variable speedup characteristics of various jobs, motivate different processor allocation strategies for multiprogramming parallel systems. Several processor allocation strategies have been presented since 1982 [11, 7, 15, 18, 12, 6, 2, 19, 3, 8, 20, 9, 14, 13, 16, 1]. The proposed strategies differ from each other in the way they allocate resources to the executing programs. Each strategy is designed to work well under certain conditions. Depending upon the system architecture, the memory organization (i.e. shared or distributed) the topology of the ....

K. Dussa, B.M. Carlson, L.W. Dowdy, K.-H. Park, "Dynamic partitioning in a transputer environment," Proc. ACM SIGMETRICS, 1990, pp. 203-213.

....terminates. In dynamic partitioning, the partitions depend on the current state of the system. The processors are reallocated every time: 1) a new job arrives and needs to be scheduled, 2) an executing job terminates, releasing its processors, or 3) the requirement of an executing job changes [3] [5], 9] 10] 13] A job does not necessarily execute in the same partition during its lifetime. In this paper, a dynamically partitioned system with non overlapping partitions is considered. This technique of multiprogramming a multiprocessor is also called dynamic space sharing [3] 9] 13] ....

....a multiprocessor is also called dynamic space sharing [3] 9] 13] In a dynamically partitioned system, partitioning can occur anytime during the execution of a job. This makes the modeling of such systems difficult. Previous efforts at modeling a dynamically partitioned system include [5], 10] In these studies, a simple system with two jobs is modeled as a closed queuing network. A Markov model for the network, based upon the possible states, is constructed and solved. Since the number of states increases significantly as the multiprogramming level is increased, the model ....

K. Dussa, B. M. Carlson, L. W. Dowdy, K. H. Park, "Dynamic partitioning in a transputer environment", in ACM SIGMETRICS , May 1990, pp. 203-213.

....[26] 1] Alternatively, preemptive policies allow executing programs to be interrupted and dynamically reallocated a larger or smaller set of processors. Examples of preemptive policies include gang scheduling [17] 4] 6] time sharing [11] 25] 9] and dynamic space sharing [9] 18] [3], 26] 12] 13] 15] Preemptive policies are optimal from an allocation point of view, since they allow for better resource utilization and can adapt to sudden changes in the workload intensity. However, the complexity of the run time environment for their implementation on an actual system ....

K. Dussa, B. Carlson, L.W. Dowdy, and K.-H. Park, "Dynamic Partitioning in a Transputer Environment," ACM SIGMETRICS, pp. 203-213, 1990.

....Oak Ridge National Laboratory managed by Martin Marietta Energy Systems, Inc. for the U.S. Department of Energy under contract no. DE AC05 84OR21400. Depending on the system architecture, preemptive or non preemptive [SRDS93] processor allocation policies have been proposed. Preemptive policies [Oust82, MEB88, PD89, ZM90, LV90, DCDP90, MVZ93, CMV94, MZ94] allow processor redistribution upon job arrivals and departures or when a time quantum expires. Processors may be reclaimed from an executing job s assignment and distributed to newly arrived jobs, or additional processors may be added to an executing job s assignment when processors become ....

K. Dussa, B.M. Carlson, L.W. Dowdy, K.-H. Park, "Dynamic partitioning in a transputer environment," Proc. ACM SIGMETRICS, 1990, pp. 203-213.

No context found.

K. Dussa, B. Carlson, L. Dowdy, and K.-H. Park, "Dynamic Partitioning in a Transputer Environment", Proceedings of the ACM SIGMETRICS,

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