D.-K. Chen, "MaxPar: an execution driven simulator for studying parallel systems," M.S. thesis, Department of Computer Science, University of Illinois, Urbana, IL, CSRD Report 917, October 1989. 153  Home/Search   Document Details and Download   Summary   Related Articles   Check

....In practice, however, large parallelism implies large overhead. This is a tradeoff faced by any parallel programmer. Consequently, measuring the parallelism of the parallel program becomes one of the main concerns in monitoring parallel software (see for example [Kuck74; Nicolau84; Kumar88; Chen89] This helps the designer know which part of the program can effectively exploit the processors, and which part is the bottleneck. Kuck et al. Kuck74] report some early measurements of parallelism in ordinary FORTRAN programs. They statically analyze the programs, and then determine which ....

....parallelism is also reported. From the profiles, some useful phases can be identified (e.g. sequential parts, parallel computations) The phases are helpful in performance tuning, performance prediction, and workload characterization. The idea introduced in [Kumar88] is used and extended in [Chen89] which describes the MaxPar execution driven simulator used to measure and identify the inherent parallelism in real application programs. The concept of dependences is described in that paper, which is useful in measuring the inherent parallelism of programs. Several techniques for detecting ....

Chen, D. -K., "MaxPar: An execution driven simulator for studying parallel systems," Master's Thesis, University of Illinois at Urbana-Champaign, 1989.

....as renaming or expansion. 4. 1 Overview of the Evaluation Method The results of the dynamic evaluation of dependence analysis reported in this paper were one of many results obtained as part of a compiler evaluation project we have been conducting at Illinois [16] as an extension of previous work [17]. The purpose of this study is to evaluate parallelizing compilers and parallelization techniques in a machine independent form. By measuring the performance of a particular program on a single machine, it is easy to determine which of two compilers is more effective. All we have to do is to ....

....longest path in the directed graph corresponding to an execution of the program under study. The ratio of the number of operations executed to the length of the critical path is the value we use for speedup. This approach was originally introduced by Kumar [19] and later extended by Chen and Yew [17] with the purpose of measuring important characteristics of sequential programs. S1: A=C 1 DO I=1,3 S2: J = J 1 S3: IF J 20 THEN S4: B(I) A B(I) ENDIF S5: A = J C ENDDO Figure 2: Code fragment for dataflow graph. To illustrate this idea, we will use the code fragment listed in ....

D.-K. Chen, "MaxPar: An Execution Driven Simulator for Studying Parallel Systems," Master's thesis, Univ. of Illinois at Urbana-Champaign, Center for Supercomputing Res. & Dev., October 1989.

....during workload execution. Other interesting techniques used to study systems include using the ECC bits of main memory to track cache misses as was done by the Wisconsin Wind Tunnel [Reinhardt et al. 1993] interrupt based profiling such as prof [Silicon Graphics] and on chip counters [Chen et al. 1995]. These techniques provide a way to study some details of a workload running on the system but lack the flexibility of complete system simulation. 6.2 High Speed Simulation The SimOS approach of using high speed instruction set simulation to study a system s behavior has been used by a number ....

CHEN, D.-K. 1995. MaxPar: An execution driven simulator for studying parallel systems.

.... all from the inner loops of several subroutines of the Eispack (a set of mathematics library routines developed at Argonne National Laboratory to solve eigenvalues and eigenvectors [20] We randomly chose one test driver routine chtest.f and used a source level performance analysis tool, MaxPar [4, 5], to obtain loop carried flow dependence information. 4 A dependence graph from each inner loop was constructed from its intermediate form by an experimental optimizing compiler developed under EPG sim environment [17] and was augmented with the loop carried dependence information mentioned ....

D.-K. Chen. MaxPar: An execution driven simulator for studying parallel systems. Master's thesis, University of Illinois at Urbana-Champaign, October 1989. Also available as CSRD tech report no. 917.

.... the inner loops of several subroutines of the Eispack (a set of mathematics library routines developed at Argonne National Laboratory to solve eigenvalues and eigenvectors [SBD 76] We randomly chose one test driver routine chtest.f and used a sourcelevel performance analysis tool, MaxPar [Che89] to obtain loop carried flow dependence information. 2 A dependence graph from each inner loop was constructed from its intermediate form by an experimental optimizing compiler developed under EPG sim environment [PY93] and was augmented with the loop2 We assume the anti and ....

D.-K. Chen. MaxPar: An execution driven simulator for studying parallel systems. Master's thesis, University of Illinois at Urbana-Champaign, October 1989. Also available as CSRD Report No. 917.

....parallelizing compiler which can target toward different granularities and machine organizations to be studied. Many of those parallelizing techniques are still at the research stage and are not available for use. To avoid these problems, we use a technique similar to [16] and [10] in MaxPar [5]. Instead of writing a simulator for a target system, simulation instructions are actually instrumented into application programs. The instrumented application programs are then executed with the simulation instructions to simulate and collect statistics for the target parallel system. This ....

....in Figure 2 statement 3 can be executed after the condition in statement 2 is resolved. But if the true branch is taken, MaxPar will allow statement 4 to be executed before statement 2 because there is no data dependence from statement 3 to statement 4 (for more details, refer to [16] and [5]) Without loss of generality, MaxPar assumes 1 time unit for each operation. All other timing parameters (e.g. synchronization overhead) use that time as the base time unit. Once we know the earliest time in which each operation can be executed, we can create an execution profile indicating the ....

[Article contains additional citation context not shown here]

D.-K. Chen. MaxPar: An Execution Driven Simulator for Studying Parallel Systems. CSRD TR-917, Center for Supercomputing Research and Development, Univ. of Illinois at Urbana-Champaign, Sep. 1989.

No context found.

D.-K. Chen, "MaxPar: an execution driven simulator for studying parallel systems," M.S. thesis, Department of Computer Science, University of Illinois, Urbana, IL, CSRD Report 917, October 1989. 153

No context found.

D.-K. Chen, "MaxPar: An Execution Driven Simulator for Studying Parallel Systems," Master's thesis, Univ. of Illinois at UrbanaChampaign, Oct. 1989.

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