M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of our symbolic analysis techniques. 1 Introduction Many parallelizing compilers and performance tools fail to effectively parallelize and evaluate programs caused by the deficit to analyze linear and non linear symbolic expressions (expressions with program unknowns) Numerous researchers [3, 20, 11, 4, 22, 14, 15, 6] have reported on the occurrence of symbolic expressions in practical codes and the need of effective techniques to analyze such programs. Non linear symbolic expressions are commonly caused by induction variable substitution, linearizing arrays, parameterizing parallel programs with symbolic ....

....only three loop nests, it makes a big difference in the overall execution time ranging by a factor of roughly 2 3 for the problem sizes measured and consistently improves for increasing problem sizes. 8 Conclusions Numerous researchers have shown the importance of symbolic compiler analysis [14, 3, 17, 15, 4, 20, 21, 11, 22, 6] for optimizing and analyzing performance of parallel programs. A crucial problem of many symbolic compiler analyses is to determine the relationship between symbolic expressions. We have described an algorithm for computing lower and upper bounds of wide classes of linear and nonlinear symbolic ....

M.J. Clement and M.J. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....as opposed to most hardware cache simulators where reference strings are generated at run time. Data reference strings are then used by a simulator whose results are less accurate than hardware simulation. However, their approach appears to be effective enough for loop optimization techniques. In [11, 10] M. Clement et al. present a compiler generated analytical model for the prediction of cache behavior, CPU execution time, and message passing overhead for scalable algorithms implemented in high level data parallel languages. The performance prediction requires a single instrumentation run of the ....

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....can be obtained from source code) seem too optimistic. Their technique was also limited to the case where parallelism is already explicit in the source program, and they did not originally conduct any study of scalability under variations in problem size. More recently, they extended this work [15] to study scalability of both the problem size and the number of processors, and build a symbolic model that represents the predicted execution time as a function of those parameters. The derivation of this model, however, required statistical methods and several experimental runs of the program ....

.... of processors (P ) and problem size (N ) like in [1] 22] and [50] Others provide a symbolic model that can be evaluated at desired combinations of N and P , but either have a very limited application domain, as in [64] or require several executions of the program for model calibration, as in [15]. There has been no proposed method, so far, that provides a first order, easily derivable model of the application s execution time (and of the execution times for internal code sections) as a function of the number of processors and problem size. Our symbolic scalability prediction method ....

Mark J. Clement and Michael J. Quinn. Symbolic performance prediction of scalable parallel programs. In Proceedings of the 9 th International Parallel Processing Symposium, April 1995.

....as opposed to most hardware cache simulators where reference strings are generated at run time. Data reference strings are then used by a simulator whose results are less accurate than hardware simulation. However, their approach appears to be effective enough for loop optimization techniques. In [17, 18] M. Clement et al. present a compiler generated analytical model for the prediction of cache behavior, CPU execution time, and message passing overhead for scalable algorithms implemented in high level data parallel languages. The performance prediction requires a single instrumentation run of the ....

M.J. Clement and M.J. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....work. In Section 3 we briefly describe P 3 T and its performance parameters. Section 4 reports on experimental results by using several realistic kernel codes taken from real world applications. Finally, some concluding remarks are made and future work is outlined. 2 Related Work In [3] M. Clement et al. present an analytical model for the prediction of cache behavior, CPU execution time, and message passing overhead. The performance prediction requires a single profile run of the program with a reduced problem size to generate a symbolic equation for execution time. M. Faerman ....

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....as opposed to most hardware cache simulators where reference strings are generated at run time. Data reference strings are then used by a simulator whose results are less accurate than hardware simulation. However, their approach appears to be effective enough for loop optimization techniques. In [9, 8] M. Clement et al. present a compiler generated analytical model for the prediction of cache behavior, CPU execution time, and message passing overhead for scalable algorithms implemented in high level data parallel languages. The performance prediction requires a single instrumentation run of the ....

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....architecturally parameterized cache simulator. Data reference strings are generated while parsing the source code and are then used by a simulator whose results are less accurate than hardware simulation. However, their approach appears to be effective enough for loop optimization techniques. In [9, 8] M. Clement et al. present a compiler generated analytical model for the prediction of cache behavior, CPU execution time, and message passing overhead for scalable algorithms implemented in high level data parallel languages. The performance prediction requires a single instrumentation run of the ....

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

....expressions, symbolic expression bounds, simplifying constraints, symbolic dependence testing 1. Introduction Many parallelizing compilers and performance tools fail to effectively parallelize and evaluate programs caused by the deficit to analyze symbolic expressions. Numerous researchers [3, 21, 11, 4, 23, 15, 16, 6] have reported on the occurrence of complex and even non linear symbolic expressions in practical codes and the need for effective techniques to analyze such programs. Non linear symbolic expressions are commonly caused by induction variable substitution, linearizing arrays, parameterizing ....

....50.0 100.0 150.0 200.0 without symbolic analysis with symbolic analysis Figure 3. Measured execution times of FTRVMT with and without symbolic analysis for varying problem sizes on a MEIKO CS 2 with 16 processors 8. Conclusions Numerous researchers have shown the importance of symbolic analysis [15, 3, 18, 16, 4, 21, 22, 11, 23, 6] for optimizing and analyzing performance of parallel programs. A crucial problem of many compiler analyses is to determine the relationship between symbolic expressions. We have described a novel algorithm for computing lower and upper bounds of wide classes of linear and non linear symbolic ....

M.J. Clement and M.J. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

.... of processors (P ) and problem size (N) like in [1] 5] and [10] Others provide a symbolic model that can be evaluated at desired combinations of N and P , but either have a very limited application domain, as in [11] or require several executions of the program for model calibration, as in [3]. There has been no proposed method, so far, that provides a first order, easily derivable model of the application s execution time (and of the execution times for internal code sections) as a function of the number of processors and problem size. Our symbolic scalability prediction method ....

M. J. Clement and M. J. Quinn. Symbolic performance prediction of scalable parallel programs. In Proceedings of the 9 th International Parallel Processing Symposium, April 1995.

....of Ethernet and ATM in PVM clusters report significant performance advantages for ATM [7, 12, 19] However, not all PVM applications will benefit from ATM technology. The impact of different network technologies on specific parallel applications can be determined with performance prediction tools [6], and accurate performance prediction is critical to making media selection decisions. Since the performance of a specific application using high speed switching technology largely depends upon the characteristics of the application itself, estimating the expected improvement is necessary to ....

M. J. Clement and M. J. Quinn. Symbolic performance prediction of scalable parallel programs. In Proceedings of the International Parallel Processing Symposium IPPS, April 25--28 1995.

No context found.

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

No context found.

M.J. Clement and M.J. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 1995.

No context found.

M. Clement and M. Quinn. Symbolic Performance Prediction of Scalable Parallel Programs. In Proc. of 9th International Parallel Processing Symposium, St. Barbara, CA, April 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