B. Armstrong and R. Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, Aug. 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....code generation from integer sets. Dikaiakos et al. 14] developed a tool called FAST that constructs task graphs from user annotated parallel programs, performs advanced task scheduling and then uses abstract simulation of message passing to predict performance. Finally, many researchers (e.g. [16, 9, 22]) have developed symbolic compile time techniques for estimating execution time, communication volume and other metrics. The communication and computation scaling functions available in our static task graph are very similar to the symbolic information used by these techniques, and could be ....

B. Armstrong and R. Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, Aug. 1998.

....message passing to predict performance. The PACE project [25] proposes a language and programming environment for parallel program performance prediction. Users are required to identify parallel subtasks and computation and communication patterns. Finally, Fahringer [15] Armstrong and Eigenmann [9], Mendes and Reed [23] and many others have developed symbolic compile time techniques for estimating execution time, communication volume and other metrics. The communication and computation scaling functions available in our static task graph are very similar to the symbolic information used by ....

B. Armstrong and R. Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, August 1998.

.... 3 1 Introduction The steady decrease in cost and increase in performance of commodity workstations and personal computers have made it increasingly attractive to use clusters of such systems as compute servers instead of high end parallel supercomputers [3,4,7]. For example, the Ohio Supercomputer Center is deploying a cluster comprising of 128 Pentium processors to serve the high end computing needs of its customers. Due to the rapid advance in performance of commodity computers, when such clusters are upgraded by addition of nodes, they become ....

....explain it here in order to make the paper self contained. A set of t independent tasks is to be mapped onto a system of m machines with the objective of minimizing the total completion time of the tasks. It is assumed that execution time for each task on each machine is known prior to execution [3,8,9,17] and contained within an ETC (Expected Time to Compute) matrix. Each row of the ETC matrix contains the estimated execution times for a given task on each machine. Similarly, each column of the ETC matrix consists of the estimated execution times for each task on a given machine. Thus, ETC[i,j] is ....

[Article contains additional citation context not shown here]

B. Armstrong, R. Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. Proceedings of the 1998 International Conference on Parallel Processing, Aug. 1998, pp.518-527

....the tasks and inter task communication. The PACE project [28] proposes a language and programming environment for parallel program performance prediction. Users are required to identify parallel subtasks and computation and communication patterns. Finally, Fahringer [16] Armstrong and Eigenmann [10], Mendes and Reed [26] and many others have developed symbolic compile time techniques for estimating execution time, communication volume and other metrics. The communication and computation scaling functions available in our static task graph are very similar to the symbolic information used by ....

Brian Armstrong and Rudolf Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, August 1998.

....the tasks and inter task communication. The PACE project [20] proposes a language and programming environment for parallel program performance prediction. Users are required to identify parallel subtasks and computation and communication patterns. Finally, Fahringer [12] Armstrong and Eigenmann [7], Mendes and Reed [18] and many others have developed symbolic compile time techniques for estimating execution time, communication volume and other metrics. The communication and computation scaling functions available in our static task graph are very similar to the symbolic information used by ....

Brian Armstrong and Rudolf Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, August 1998.

....using run time as an evolution criteria. The static assignment thus obtained outperformed the dynamic allocation method by nearly 20 . At Purdue University we have characterized the computation, communication and I O behavior of the SPECseis application on a number of computer architectures [2]. Based on this work we have developed a performance forecasting methodology and tool that can extract this behavior from a given application program and express it mathematically in function of the program s input data and architectural parameters (such as the number of processors and network ....

Brian Armstrong and Rudolf Eigenmann. Performance forecasting: Towards a methodology for characterizing large computational applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518--525, August 1998.

....requires that a single threshold be chosen. In contrast, our approach adapts to the architecture on which the program executes. Techniques used to serialize unprofitable loops aim at predicting the performance of a parallel loop. Much work has been done in the area of performance prediction [7, 8, 9]. 5 However, these techniques can be computationally intensive and therefore too expensive to include in a run time test. In contrast, the performance prediction approaches used in this paper are simple and make use of profiling and direct program measurements. Our technique uses a scheme ....

Brian Armstrong and Rudolf Eigenmann. Performance forecasting: Towards a methodology for characterizing large computational applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518--525, August 1998.

....using run time as an evolution criteria. The static assignment thus obtained outperformed the dynamic allocation method by nearly 20 . At Purdue University we have characterized the computation, communication and I O behavior of the SPECseis application on a number of computer architectures [AE98]. Based on this work we have developed a performance forecasting methodology and tool, which can extract this behavior from a given application program and express it mathematically in function of the program s input data and architectural parameters (such as the number of processors and network ....

Brian Armstrong and Rudolf Eigenmann. Performance forecasting: Towards a methodology for characterizing large computational applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518--525, August 1998.

....degradation and the sections of the code and data structures where they occur. The PTOPP model [11] defines overhead factors, such as globalization penalty, parallelization overhead, and spreading overhead, and recipes that guide the user in optimizing a parallel program. The Perfore methodology [1] extracts Resource Usage Equations from an applications, which can be used to characterize the scaling behavior of the application with respect to input data and architecture parameters. The presentation of the results of these methods is beyond the scope of this paper. However, it is an ....

Brian Armstrong and Rudolf Eigenmann. Performance forecasting: Towards a methodology for characterizing large computational applications. In Proceedings of the International Conference on Parallel Processing, pages 518--525, August 1998.

No context found.

B. Armstrong and R. Eigenmann. Performance Forecasting: Towards a Methodology for Characterizing Large Computational Applications. In Proc. of the Int'l Conf. on Parallel Processing, pages 518-525, Aug. 1998.

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