M. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems," in Job Scheduling Strategies for Parallel Processing (D. G. Feitelson and L. Rudolph, eds.), vol. 949 of Lecture notes in Compter Science, Springer-Verlag, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the perturbations caused by I O, that I O bound jobs suffer under gang scheduling, and therefore argue in favor a flexible gang scheduling. 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, ....

M. S. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems". In Job Scheduling Strategies for Parallel Processing, Notes in Computer Science Vol. 949.

....machines becomes a more pressing issue. Job scheduling is an important aspect of this work, and much research on the topic of job scheduling for parallel computers has been done in recent years. Some have focused on distributed memory machines [7, 8, 9, 11, 14] others on more general systems [3, 6, 10, 15]. Many different strategies (static vs. dynamic, time sharing vs. space sharing, etc. have been investigated and compared. All these studies make underlying assumptions about the workload in order to quantify their results. Some of these studies use synthetic workloads 1 whose characteristics ....

M.S. Squillante. "On the Benefits and Limitations of Dynamic Partitioning in Parallel Computer Systems". IPPS'95 Workshop on Job Scheduling Strategies for Parallel Processing, April, 1995.

.... 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 ....

....This avoids the inefficiencies of executing the job on an unsuitable number of PEs just to see what happens. One problem with dynamic partitioning is the overhead of re partitioning, and possible need to transfer computation state from one processor to another (in distributed memory machines) [446, 441, 543, 289]. The common solution is to limit the rate of such reallocations [422] or to perform them at predefined points in the application where the state is minimal [630, 629, 185, 521] In addition, it should be noted that proposed dynamic partitioning schemes suffer from the same memory oblivion as ....

M. S. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 219--238, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

.... 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 ....

M. S. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 219--238, Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....(used by MPI) and dataparallel (used by HPF) models. Moreover, it could lead to extensive queueing if sufficient processors are not available. And the overheads for repartitioning, which include switching processors from one protection domain to another, may negate the benefits that were expected [44]. In addition, dynamic partitioning requires extensive coordination between the operating system and the application s internal scheduler, so as to handle the changes in processor allocation efficiently [2] For example, if a processor is taken away from an application, the whole application ....

M. S. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

.... jobs, and avoiding the need for non work conserving algorithms [21] Several studies have revealed that EQUI 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 ....

M. S. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems". In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp. 219--238, SpringerVerlag, 1995. Lecture Notes in Computer Science Vol. 949.

....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 ....

M. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems . In D. Feitelson and L. Rudolph, editors, Job scheduling strategies for parallel processing, volume 949 of Lecture Notes in Computer Science. Springer-Verlag, 1995.

....re submitted to the system, and computation would continue from the point where the job was forced to vacate the machine. In this way, the time already invested in the job will be preserved. Another use for checkpointing in a parallel system is to perform dynamic partitioning which has been shown [2] to be more effective than static methods of scheduling parallel programs. In a dynamic partitioning scheme, the number of resources allocated to a job is changed while the job is running based on changes in load on the overall system. Without the ability to checkpoint and save the state of ....

M. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems, " in Job Scheduling Strategies for Parallel Processing (D. G. Feitelson and L. Rudolph, eds.), vol. 949 of Lecture notes in Compter Science, Springer-Verlag, 1995.

....[18] Dynamic policies introduce costs related to data and job migration, processor preemption and coordination, and application reconfiguration. Their performance and implementation feasibility are a function of the underlying architecture and programming model as costs may outweigh benefits [29] (e.g. like in distributed memory systems) Dynamic policies that limit the number of preemptions strive to accommodate such problems [3] We conclude the discussion on processor only scheduling by discussing gang scheduling, which effectively integrates space sharing and time sharing. In ....

....and time sharing. In Ousterhout s seminal paper [20] on coschedul ing the issue of integrating processor allocation with process dispatching (i.e. both levels of scheduling) was proposed. Gang scheduling ensures that all running processes of a parallel job execute in the same time quantum [20, 16, 29] and it is a particularly useful scheme for guaranteeing good interactive response times. 3.1.3 Considerations for Dual Resource Allocation Policies Another important aspect in scheduling is that of coordinating the allocation of dual resources, thus making a hard problem even harder. Given that ....

Squillante M.S., "On the Benefits and Limitations of Dynamic Partitioning in Parallel Computer Systems" in Feitelson D.G., Rudolph L., (Eds.), Job Scheduling Strategies for Parallel Processing, Springer-Verlag, Lecture Notes in Computer Science, Vol 949, pp 219-238, 1995.

....real parallel systems executing real scientific and or commercial workloads. An exact solution of the model is obtained by employing matrix geometric techniques [24] In this paper we provide a less formal and rigorous description of our mathematical analysis, and we refer the interested reader to [37, 38] for additional technical details. It is important to note that the computational efficiency of our approach allows us to examine the large design space of diverse parallel environments. We use this modeling framework to consider the fundamental question: how expensive must the costs of ....

.... and both an arrival and a departure within a small time interval are all assumed to occur with negligible probability, leading to a quasi birth death process [24] although our analysis is easily extended to handle batch arrivals and or departures as long as the batch sizes are bounded; see [38]) The use of phase type distributions [24] for the parameters of our model is motivated in part by their important mathematical properties, which can be exploited to obtain a tractable analytic model while capturing the fundamental aspects of dynamic partitioning. Just as important, however, is ....

[Article contains additional citation context not shown here]

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. Tech. Rep. RC

....The technical details of our models and their solutions are beyond the scope of this paper. We refer the interested reader to [35] for derivations of an exact solution of each model, including expressions for performance measures of interest. Additional details on the models can be found in [35, 36]. 4.1 Dynamic Partitioning We model a parallel computer system consisting of P identical nodes that are scheduled according to the (basic) DEP policy defined in Section 2.1. Recall that the node allocations are reconfigured whenever a job 5 0 1 2 3 4 5 6 7 8 Number of Nodes 0 1 2 3 4 5 ....

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.). Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....especially those with lower efficiencies and or those with less variable service time requirements [45, 49, 14, 26, 31, 27] This is because the dynamic policy can maintain very efficient node utilizations by adjusting node allocation according to workload changes. On the other hand, studies [31, 40, 18] have demonstrated that the overheads associated with dynamic partitioning in distributed environments can generally limit and or eliminate its potential performance benefits (due to factors such as data job migration, node preemption coordination, and reconfiguration of the application) and that ....

....and thus it can be particularly suitable for tasks with small processing requirements. It can also provide the best system performance in the limiting extreme of linear speedup workloads, as all node allocations are equally efficient and the fully parallel allocation reduces job execution time [40]. On the other hand, it is most often the case that the speedup of parallel jobs does not increase linearly with the number of processors allocated to the job [7, 23] which results in poor utilization of the resources in time sharing systems [23, 43, 44, 47] Moreover, the context switching ....

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pages 219--238. Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....can maintain very efficient node utilizations by adjusting node allocation according to workload changes. However, for other application workloads, the overheads associated with greedy dynamic partitioning in distributed environments can generally eliminate its potential performance benefits [7, 2]. In this section we describe a new approach for dynamically adjusting the partitioning of resources at any level of our resource scheduling system hierarchy in a controllable manner while reducing the overheads associated with dynamic partitioning in large scale parallel environments. One dynamic ....

....node allocations, which can be set by the system administrator and can also be dynamically adjusted. This approach reduces the number of reconfigurations, decreases the total costs of repartitioning by performing multiple reconfigurations together, and prevents a form of reconfiguration thrashing [7]. It also tends to reduce the adverse effects of highly variable job arrivals, and tends to result in better system performance by batching multiple allocation decisions. Our resource management system uses this DPS approach to determine the size of each partition P d p comprising every domain D ....

M. S. Squillante, On the benefits and limitations of dynamic partitioning in parallel computer systems, in Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), Springer-Verlag, 1995, pp. 219--238. Lecture Notes in Comp. Sci. Vol. 949.

....The technical details of our models and their solutions are beyond the scope of this paper. We refer the interested reader to [37] for derivations of an exact solution of each model, including the calculation of performance measures of interest. Additional details on the models can be found in [37, 38]. 4.1 Dynamic Partitioning We model a parallel computer system consisting of P identical nodes that are scheduled according to the (basic) dynamic equi partitioning policy defined in Section 2.1. Recall that the node allocations are reconfigured whenever a job arrives to a system with 0 # i N ....

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.). Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....combine space sharing and time sharing, in what is often called gang scheduling [15, 4] provide the benefits of both space sharing and time sharing. Gang Scheduling can be effective with small interactive jobs, and can reduce the mean job response time (for a given partitioning of the nodes) [16]. These advantages are very important especially under highly variable workloads although they come with the price of increased overhead for context switching among the parallel jobs and fragmentation of node allocations. Loosely synchronized applications are applications with coarse grained ....

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pp 219--238. Springer-Verlag, 1995. Lecture Notes Comp. Sci. Vol. 949.

....processors a priori into partition pools belonging to each job class. Second, the partition size of a large or medium job can be dynamically adjusted during its execution by suitably reconfiguring the application. Small jobs are not dynamically reconfigured because of overhead considerations [35]. The operation of the DP policy is controlled via the seven parameters shown in Table I. The system maintains a single FCFS queue for waiting jobs, and a list of running jobs that are eligible for reconfiguration, which we call the interruptible list. After a medium or large job reconfigures ....

....interruptible list. After a medium or large job reconfigures itself, the job is removed from this list and is not eligible for reconfiguration until a certain time interval, denoted by T , has elapsed. This mechanism is motivated in part by the need to prevent a form of reconfiguration thrashing [35]. When there are no other constraints, the DP policy allocates a valid processor partition to a new job such that the size of this partition is close to the average size of all currently allocated partitions in the system (including this new partition) Partition reconfigurations are also guided ....

[Article contains additional citation context not shown here]

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pages 219--238. Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

....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 [6,15,18,25,27 30,43,46,50,53]. Gang scheduling can be viewed as a hybrid scheduling scheme that combines time sharing with space sharing [9 11,33] Under this scheduling approach, jobs with the same resource requirements are collected in groups, and each group is assigned to a partition of the system for a specific amount ....

M. S. Squillante. On the benefits and limitations of dynamic partitioning in parallel computer systems. In Job Scheduling Strategies for Parallel Processing, D. G. Feitelson and L. Rudolph (eds.), pages 219--238. Springer-Verlag, 1995. Lecture Notes in Computer Science Vol. 949.

No context found.

M. Squillante, "On the benefits and limitations of dynamic partitioning in parallel computer systems," in Job Scheduling Strategies for Parallel Processing (D. G. Feitelson and L. Rudolph, eds.), vol. 949 of Lecture notes in Compter Science, Springer-Verlag, 1995.

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