Eager DL, Zahorjan J, Lazowska ED. Speedup versus efficiency in parallel systems. IEEE Transactions on Computers 1983; 38(3):408--423.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....about the workload is available a priori, and that the number of processors requested by applications is fixed during the complete execution, and it is known at submission time. Other works determine the number of processors that must be allocated to applications as a function of their speedups [Eager89] [Ghosal91] but having the complete speedup curve as input. Our research is based on execution environments where no knowledge of the applications is provided a priori, and it is focused on determining the number of processors that must be allocated to each application. In this work, we assume ....

D. L. Eager, J. Zahorjan,E.D.Lawoska. "Speedup Versus Efficiency in Parallel Systems", IEEE Trans. on Comp., vol. 38,(3), pp. 408-423, March 1989.

....of the fact that massively parallel, distributed systems are an attractive alternative to supercomputers in terms of both price and performance for many applications. One of the outcomes of research in such scalable systems has been the development of various models of performance. For example, [9], 13] and [15] discuss various aspects of speedup and efficiency of parallel systems. Although insightful, these do not provide explicit analytical models. One goal of our research is to identify an analytical model that predicts the execution time of an application given a set of parameters for ....

D. Eager, J. Zahorjan, and E. Lazowska. Speedup versus efficiency in parallel systems. IEEE Trans. Computs., 38(3), 1989.

....et al. 29] propose several processor allocation schemes based on the processor working set (PWS) which is the number of allocated processors for which the ratio of execution time to efficiency is minimized. The PWS differs from the average parallelism of the job by at most a factor of two [16]. The best of the variants of PWS gives jobs at most their processor working set, but under heavy load gives fewer and fewer processors to each job, thus increasing efficiency and therefore system capacity. Setia, Squillante, and Tripathi [74] use a queuing theoretic model to investigate how ....

....based on knowledge of the processing requirement (such as least work first) were still better. Knowledge of the average parallelism of a job makes it possible to allocate each job an appropriate number of processors to make it operate at a near optimal ratio of execution time to efficiency [16]. With the knowledge of how many processors each job uses, policies for packing the jobs into frames for gang scheduling are investigated by Feitelson [18] Feitelson and Rudolph [22] describe a discipline in which processes that communicate frequently are identified, and it is assured that the ....

D. L. Eager, J. Zahorjan, and E. D. Lazowska, "Speedup versus efficiency in parallel systems". IEEE Trans. Comput. 38(3), pp. 408--423, Mar 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 . No load balancing This is obviously the worst case. Each gang is mapped on the same subtree in which it ....

Eager, D. L., Zahorjan, J., and Lazowska, E. D. Speedup versus efficiency in parallel systems. IEEE Trans. Comput. 38, 3 (Mar. 1989), 408--423.

.... is that s effic represents the knee in the speed up curve, i.e. the point that maximizes the benefit cost ratio (benefit meaning lower execution time and cost meaning use of more processors ) This concept is in consonance with a more formal analysis of speedup behavior by Eager et al. [36]. Eager et al. found that (i) the knee k of the speed up curve must satisfy kA , and (ii) adding more processors when the partition size is smaller than A has much greater impact on the execution time than when the par 46 tition size is greater than A [36] These results provide the ....

.... of speedup behavior by Eager et al. [36] Eager et al. found that (i) the knee k of the speed up curve must satisfy kA , and (ii) adding more processors when the partition size is smaller than A has much greater impact on the execution time than when the par 46 tition size is greater than A [36]. These results provide the rationale for modeling A after the efficient partition size s effic . 1 4 5 10 11 30 31 50 51 100 100 Don t Know 0 5 10 15 20 25 30 35 40 of Answers Figure 32 Survey results for efficient partition size s effic Figure 32 displays how the efficient partition ....

Derek Eager, John Zahorjan, and Edward Lazowska. Speedup Versus Efficiency in Parallel Systems. IEEE Transactions on Computers, vol. 38, no. 3, March 1989.

....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 ....

....and maximum 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 ....

D. Eager, J. Zahorjan, and E. Lazowska. "Speedup Versus Efficiency in Parallel Systems". IEEE Transactions on Computers 38(3), March 1989.

....about the workload is available a priori, and that the number of processors requested by applications is fixed during the complete execution, and it is known at submission time. Some works determine the number of processors that must be allocated to applications as a function of their speedups [Eager89] [Ghosal91] but having the complete speedup curve as input. Our research is based on execution environments where no knowledge of the applications is provided a priori, and it is focused on determining the number of processors that must be allocated to each application. In this work, we assume ....

D. L. Eager, J. Zahorjan,E.D.Lawoska. "Speedup Versus Efficiency in Parallel Systems ", IEEE Transactions on Computers, vol. 38,(3), pp. 408-423, March 1989.

....the decisions. In order to calculate the efficiency, they measure the different sources of overheads that cause loss of efficiency and subtract them from 1. Figure 1 shows the formulation proposed by Nguyen in [10] to calculate the efficiency and the relationship between efficiency and speedup[4]. The sources of overhead, showed in Figure 1, are the system overhead, the idleness and communication (processor stall time) 2 .These components were obtained by using the hardware counters provided by the architecture, and by instrumenting the parallel library. However, one of the major ....

D. L. Eager, J. Zahorjan and E. Lazowska, "Speedup Versus Efficiency in Parallel Systems". IEEE TRans. Computer. 38 (3) 1989 pp 408-423

....performance tools, and empirical performance models, all provide ways to analyze the scalability of applications on DSM machines. Mathematical models have been widely used. For example, they have been used to model the effect of load imbalance [5] the tradeoff between speedup and efficiency [4], and the contention in shared memory machines [6] While they are fast, they use simplified models, often with assumptions that restrict their accuracy and their applicability to real machines. Software based simulators simulate simplified machine models in detail. They tend to be more accurate ....

D. Eager et al. Speedup versus efficiency in parallel systems. IEEE Transactions on Computers, 38(1), 1989.

.... 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 gang scheduling. ....

....vector of length l. Allocating the Theta bounding box results violates property (1) One proposed solution to this problem is to allocate a number of processors equal to the average vector length (instead of the maximum vector length) and to fold any longer vectors into multiple segments [EZL89, MEB91] But this method can be shown to violate property (2) if vector lengths change unpredictably during execution. Related work may be found in [Iye95] and [Ble90, Chapter 12] 95 R1 R2 R4 R1 R3 R4 R6 R7 R9 Sequential Memory Data Parallel Memory Figure 5.1. Dynamic allocation ....

D.L. Eager, J. Zahorjan, and E.D. Lazowska. Speedup versus efficiency in parallel systems. IEEE Transactions on Computers, C-38(3):408-- 423, 1989.

.... the cost function C(p) is defined as C(p) p S p and thus j(p) S p 2 p The pws is the minimum value of p which maximizes j(p) It is shown that for the cost function used, pws is the same as the number of processors associated with the knee of the execution time efficiency profile [EZL89] Some adaptive policies utilizing pws were investigated. Among them, the best power was observed with a simple strategy of allocating a job s pws when possible. If no jobs in the queue can be allocated their pws, the free processors are allocated to the first job in the queue. This has the ....

Derek L. Eager, John Zahorjan, and Edward D. Lazowska. Speedup versus Efficiency in Parallel Systems. IEEE Transactions on Computers, 38(3):408--23, March 1989.

....both, the use of hardware counters and the measurement of the idle periods of the applications. The most studied characteristic of parallel applications has been the speedup. Several theoretical studies have analyzed the relation between the speedup and other characteristics such as the efficiency [8]. Helmbold et al. analyze in [11] the causes of loss of speedup and demonstrate that the super lineal speedup exists basically due to memory cache effects. Our work has several characteristics that differ from the previously mentioned proposals. First of all, with respect the parameters used by ....

D.L.Eager,John Zahorjan,E.D.Lawoska. "Speedup Versus Efficiency in Parallel Systems", IEEE Transactions on Computers,Vol. 38,(3), pp 408-423, March 1989

No context found.

Eager DL, Zahorjan J, Lazowska ED. Speedup versus efficiency in parallel systems. IEEE Transactions on Computers 1983; 38(3):408--423.

No context found.

D. L. Eager, J. Zahorjan, and E. D. Lazowska. Speedup versus efficiency in parallel systems. IEEE Trans. on Computers, 38(3):408--423, March 1989.

No context found.

D. Eager, J. Zahorjan, and E. Lazowska. Speedup versus Efficiency in Parallel Systems. IEEE Transactions on Computers, 38(3):408 -- 423, March 1989.

No context found.

D. Eager, J. Zahorjan, and E. Lazowska. Speedup versus efficiency in parallel systems. IEEE Transactions on Computers, 38(3), 408--423, March 1989.

No context found.

D. Eager, J. Zahorjan, and E. Lazowska. Speedup versus Efficiency in Parallel Systems. IEEE Transactions on Computers, 38(3):408 -- 423, March 1989.

No context found.

D. L. Eager, J. Zahorjan, and E. D. Lazowska, "Speedup Versus Efficiency in Parallel Systems," IEEE Transactions on Computers, Vol.38, No.3, March 1989, pp.408--423.

No context found.

D. L. Eager, J. Zahorjan, E. D. Lawoska. "Speedup Versus Efficiency in Parallel Systems", IEEE Trans. on Computers, Vol. 38,(3), pp. 408-423, March 1989.

No context found.

D. L. Eager, J. Zahorjan, and E. D. Lazowska. Speedup versus efficiency in parallel systems. IEEE Trans. on Computers, 38(3):408--423, March 1989.

No context found.

D. Eager, J. Zahorjan, and E. Lazowska, "Speedup versus efficiency in parallel systems," IEEE Transactions on Computers, vol. 38, no. 3, pp. 408--423, March 1989.

No context found.

D.L. Eager, J. Zahorjan, and E.D. Lazowska, Speedup versus Efficiency in Parallel Systems, IEEE Trans. Computers, vol. 38, no. 3, pp. 408423, Mar. 1989.

No context found.

D. L. Eager, J. Zahorjan, E. D. Lawoska, "Speedup Versus Efficiency in Parallel Systems", IEEE Transactions on Computers, vol. 38,(3), pp. 408-423, March 1989.

No context found.

DL Eager, J Zahorhan, and ED Lazowska, "Speedup Versus Efficiency in Parallel Systems", IEEE Transactions on Computers 38(3), pp. 408-423, March 1989.

No context found.

D. L. Eager, J. Zahorjan, E. D. Lawoska, "Speedup Versus Efficiency in Parallel Systems ", IEEE Transactions on Computers, vol. 38,(3), pp. 408-423, March 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