C.Yu and C.R.Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes," Proc. Int. Conf. Parallel Processing, pp. II-143-150, 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....topological 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 ....

....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 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 ....

C.Yu and C.R.Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes ", Proc. Int. Conf. Parallel Processing, pp.II143 -150, 1994.

....fact, equipartition itself is an ideal, that is only approximated by dynamic partitioning. In the context of adaptive partitioning, partition sizes cannot be changed after the initial allocation, even if new jobs continue to arrive. The algorithm for equipartition in this framework is as follows [514, 628]. As each job arrives, there either are free PEs available or not. If there are, the job is allocated PEs according to its request (or all of them, if less are available) If there are no free PEs, the job is queued. When a job terminates, its PEs are divided equally among all the jobs in the ....

....some jobs will be left in the queue. A few other algorithms have been devised that are not derived directly from the equipartition ideal. These algorithms explicitly try to reserve PEs for future arrivals. The first algorithm simply limits the maximal partition size that any single job can obtain [489, 628, 490]. Thus no single job can monopolize the whole system to the detriment of others. It has been 34 shown that adaptive partitioning with a maximal partition size leads to better overall efficiency than variable partitioning based on the average parallelism in applications, provided the maximum is ....

[Article contains additional citation context not shown here]

C. Yu and C. R. Das, "Limit allocation: an efficient processor management scheme for hypercubes". In Intl. Conf. Parallel Processing, vol. II, pp. 143--150, Aug 1994.

....fact, equipartition itself is an ideal, that is only approximated by dynamic partitioning. In the context of adaptive partitioning, partition sizes cannot be changed after the initial allocation, even if new jobs continue to arrive. The algorithm for equipartition in this framework is as follows [309, 363]. As each job arrives, there either are free PEs available or not. If there are, the job is allocated PEs according to its request (or all of them, if less are available) If there are no free PEs, the job is queued. When a job terminates, its PEs are divided equally among all the jobs in the ....

.... only at the price of reduced efficiency [103, 349, 72, 128, 71, 199] This is a problem for scheduling schemes based on static partitioning (including and gang scheduling) Adaptive and dynamic partitioning, on the other hand, only gives a job more PEs if there is nothing better to do with them [199, 363, 312, 152, 259, 239]. When multiple jobs compete for resources, each is limited to a relatively small number of PEs, thus boosting its efficiency. Effect on utilization and throughput In uniprocessors, multiprogramming improves utilization by overlapping the computation of one job with the I O of another, using time ....

C. Yu and C. R. Das, "Limit allocation: an efficient processor management scheme for hypercubes". In Intl. Conf. Parallel Processing, vol. II, pp. 143--150, Aug 1994.

....assign these nodes to the waiting job(s) Allocation algorithms with better recognition ability can improve the chance of assigning a job into the system. Although the recognition ability varies dramatically among allocation algorithms, the actual performance does not show significant difference [12, 13]. This is because of the fragmentation problem. Fragmentation occurs when the required number of nodes are available but they do not form the topology required for task execution. The performance gain observed from efficient allocation schemes is limited mainly by the first come firstserve (FCFS) ....

....with most allocation algorithms or in both directions when the two dimensional buddy algorithm is used. Folding a job evenly ensures that the workload of the processors are increased evenly and no bottleneck is introduced. A job can be folded further if necessary. Unlike the limit allocation [13] which limits the largest executable job size, to ensure the reduced waiting time is not overwhelmed by the increased execution time caused by job folding, we restrict the number of times a job can be folded. The method is thus called the restricted size reduction (RSR) method. The RSR method is ....

[Article contains additional citation context not shown here]

C. Yu and C. R. Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes, " Int. Conf. on Parallel Processing, vol. II, pp. 143-150, 1994.

....to detect free nodes and to decide whether the free nodes can form a submesh for the execution of a job. Allocation algorithms with better recognition ability for available submeshes can improve the chance of assigning a job into the system and reduce the waiting delay. However, as studies [8, 9] showed, significant performance improvement cannot be obtained by refining the conventional allocation algorithms. A multicomputer is typically underutilized when operated in a dynamic environment because of fragmentation problems. Fragmentation occurs when there are free nodes in the system but ....

....list of free submeshes for the allocation. Its allocation process has a lower time complexity than that of the busy list scheme but its deallocation time complexity is higher. Its submesh recognition ability is similar to that of the adaptive scan and busy list algorithms. Limit: Limit allocation [9] was proposed for the hypercube system and later extended to the mesh topology in two independent works [10, 12] The idea is to allocate a job to a smaller size submesh if the required submesh cannot be allocated. Probability of finding a smaller size submesh is larger and therefore the average ....

[Article contains additional citation context not shown here]

C. Yu and C. R. Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes," Int. Conf. on Parallel Processing, vol. II, pp. 143-150, 1994.

....jobs per each random number seed. With 20 seeds, we observe the first 20,000 jobs, which is sufficient to obtain the steady state parameters. We measure the mean response time which includes the queueing delay as well as the execution (service) time. For the detailed simulation environment, refer [5]. Figure 6 shows the variation of mean response time with respect to input load or system utilization for a 4 ary 3 cube (Q or 22 x 22 x 22) The Isomorphic allocation algorithm (we call the Semi Isomorphic Strategy as the Isomorphic since there is no ambiguity here) outperforms other policies ....

C.Yu and C.R.Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes," Proc. Int. Conf. Parallel Processing, pp. II-143-150, 1994.

No context found.

C. Yu and C. R. Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes, " Proc. Int'l Conf. on Parallel Processing, Vol. II, pp. 143-150, Aug. 1994.

No context found.

C.Yu and C.R.Das, "Limit Allocation: An Efficient Processor Management Scheme for Hypercubes," submitted to ACM SIGMETRICS Conf. Meas. and Mod. of Comp. Sys. Vita Chansu Yu, the fourth son of Kuk-hyun Yu and Yok-young Wi, was born in Seoul, Korea, on February 1,

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