ipps94 (1158310), страница 4
Текст из файла (страница 4)
To associateperformance data with these important operationallysemantic components, we formulated an instrumentation model that we then applied to detailed performance studies. This instrumentation model is represented in Figure 6. Essentially, we identify events atthe begin and the end of the program as well as atthe begin and end of each major execution component(event capture points are indicated by circled numbersin the gure). From these events, the following timemeasurements are computed:(1)-(8) (setup) : The entire benchmark program.(2)-(8) (fork) : The part of the program which runsin parallel, starting with the forking of processes.(3)-(7) (main) : The main pC++ program as supplied by the user.
It includes static collection allocation, user setup and wrapup, and the parallelalgorithm.(4)-(7) (user) : The computation part representing\pure" user code execution, without the overheadof collection allocation.(5)-(6) (parallel) : The parallel algorithm portion ofthe pC++ program. The time in this section corresponds to the measurements reported in x5.Our rst application of the above instrumentationand measurement model was to determine, using tracing, the relative inuence of dierent phases of theentire benchmark execution where the language andruntime system execution components were involved.Because the speedup results reported in x5 are onlyfor the parallel section, we wanted to characterizethe speedup behavior in other program regions.
Anexample of the detailed performance information weare able to obtain is shown for the Poisson benchmark in Figures 7, 8, and 9 for the shared memorypC++ ports. In addition to the phases describedabove, the gures show also the speedup prole of thesineTransform (t) and cyclicReduction (cyclic)functions as described in Section 4.2. The graphsclearly show how overall performance is degraded bycomponents of the execution other than the main parallel algorithm, which scales quite nicely. Althoughsome of these components will become relatively lessimportant with scaled problem sizes, understandingwhere the ineciencies lie in the pC++ execution system will allow us to concentrate optimization eortsin those areas.We use proling measurements as a way of obtaining more detailed performance data about runtime system functions, thread-level application functions, collection class methods, and collection referencing.
The pC++ proler and instrumentation toolsallow dierent levels and types of performance information to be captured. Whereas the type of measurement above helps us identify pC++ system problems,a pC++ programmer may be particularly interestedin information about where the parallel algorithms isspending the most time and its collection referencingbehavior.7 ConclusionThe pC++ programming system includes an integrated set of performance instrumentation, measurement, and analysis tools. With this support, we havebeen able to validate performance scalability claims ofthe language and characterize important performancefactors of the runtime system ports during pC++ system development. As a consequence, the rst versionof the compiler is being introduced with an extensiveset of performance experiments already documented.Some of the performance analysis has been reportedin this paper.
From the scalability data, we see thatpC++ already achieves good performance. From thedetailed trace and prole data we are able to pinpointthose aspects in the language's use for algorithm design or in the implementation of runtime system operations where performance optimizations are possible.For instance, the proler has shown that a great number of barrier synchronization are generated, causing areduction in overall parallelism. One important compiler optimization will be to recognize when barrierscan be removed or replaced with explicit synchronization. Other optimizations might be more architecturespecic.
As an example, in distributed memory systems it will be important to overlap communicationwith computation, whereas in shared memory environments, collection distribution and memory placement will be important for achieving good locality ofreference. Again, performance analysis will be criticalfor identifying the need and resulting benet of suchoptimizations.For more information ...Technical documents and the programs for pC++and Sage++ are available via anonymous FTP frommoose.cs.indiana.edu and ftp.cica.indiana.eduin the directory ~ftp/pub/sage. We maintain twomailing lists for pC++/Sage++.
For informationabout the mailing lists, and how to join one, pleasesend mail to sage-request@cica.indiana.edu. Nosubject or body is required. Also, the pC++/Sage++project has created a World-Wide-Web server. Use aWWW viewer to browse the on-line user's guides andpapers, get programs, and even view pictures of thedevelopment team. The server can be found at http://www.cica.indiana.edu/sage/home-page.html.References[1] BBN Advanced Computer Inc., Cambridge, MA. Inside the TC2000, 1989.[2] S.
Frank, H. Burkhardt III, J. Rothnie, The KSR1:Bridging the Gap Between Shared Memory and MPPs,Proc. Compcon'93, San Francisco, 1993, pp. 285{294.[3] D. Gannon, F. Bodin, S. Srinivas, N. Sundaresan, S.Narayana, Sage++, An Object Oriented Toolkit forProgram Transformations, Technical Report, Dept. ofComputer Science, Indiana University. 1993.[4] V. Herrarte, E.
Lusk, Studying Parallel Program Behavior with Upshot, Technical Report ANL-91/15,Mathematics and Computer Science Division, ArgonneNational Laboratory, August 1991.[5] Sequent Computer Systems, Inc. Symmetry Multiprocessor Architecture Overview, 1992.[6] A. Chien and W. Dally. Concurrent Aggregates (CA),Proc. 2nd ACM Sigplan Symposium on Principles &Practice of Parallel Programming, Seattle, Washington, March, 1990.[7] High Performance Fortran Forum, High PerformanceFortran Language Specication, 1993. Available fromtitan.cs.rice.edu by anonymous ftp.[8] J.
K. Lee, Object Oriented Parallel Programming Paradigms and Environments For Supercomputers, Ph.D.Thesis, Indiana University, Bloomington, Indiana, Jun1992.[9] J. K. Lee, D. Gannon, Object Oriented Parallel Programming: Experiments and Results, Proc. Supercomputing 91, Albuquerque, IEEE Computer Society andACM SIGARCH, 1991, pp. 273{282.[10] D. Gannon, J.
K. Lee, Object Oriented Parallelism:pC++ Ideas and Experiments, Proc. of 1991 Japan Society for Parallel Processing, pp. 13{23.[11] D. Gannon, J. K. Lee, On Using Object Oriented Parallel Programming to Build Distributed Algebraic Abstractions, Proc. CONPAR 92{VAPP, Lyon, France,Sept. 1992.[12] D. Gannon, Libraries and Tools for Object ParallelProgramming, Proc. CNRS-NSF Workshop on Environments and Tools For Parallel Scientic Computing,St. Hilaire du Touvet, France, Elsevier, Advances inParallel Computing, Vol.
6, pp. 231{246, 1993.[13] B. Mohr, Performance Evaluation of Parallel Programs in Parallel and Distributed Systems, H. Burkhart (Eds.), Proc. CONPAR 90{VAPP IV, Joint International Conference on Vector and Parallel Processing,Zurich, Lecture Notes in Computer Science 457, pp.176{187, Berlin, Heidelberg, New York, London, Paris,Tokio, Springer Verlag, 1990.[14] B. Mohr, Standardization of Event Traces Considered Harmful or Is an Implementation of ObjectIndependent Event Trace Monitoring and Analysis Systems Possible?, Proc. CNRS-NSF Workshop on Environments and Tools For Parallel Scientic Computing,St.
Hilaire du Touvet, France, Elsevier, Advances inParallel Computing, Vol. 6, pp. 103{124, 1993.[15] J. K. Ousterhout, Tcl: An Embeddable CommandLanguage, Proc. 1990 Winter USENIX Conference.[16] J. K. Ousterhout, An X11 Toolkit Based on the TclLanguage, Proc. 1991 Winter USENIX Conference.[17] F. Bodin, P. Beckman, D. Gannon, S. Yang, S. Kesavan, A. Malony, B. Mohr, Implementing a Parallel C++ Runtime System for Scalable Parallel Systems, Proc. 1993 Supercomputing Conference, Portland, Oregon, pp.
588{597, Nov. 1993.[18] D. A. Reed, R. D. Olson, R. A. Aydt, T. M. Madhyasta, T. Birkett, D. W. Jensen, B. A. A. Nazief,B. K. Totty, Scalable Performance Environments forParallel Systems. Proc. 6th Distributed Memory Computing Conference, IEEE Computer Society Press, pp.562{569, 1991.[19] J. Palmer, G. L.
Steele, Jr. Connection MachineModel CM-5 System Overview, Proc, 4th Symp. Frontiers of Massively Parallel Computation, pp. 474{483.[20] Intel Supercomputes, Paragon-XP/S Technical Specication., Beaverton, Or.(256.0) 4.0EmbarGridPoissonSparseEmbarGridPoissonSparse20.03.015.010.02.05.0(64.0) 1.0050100150200250300Figure 1: Benchmark Speedups for TMC CM-50.001020Figure 3: Benchmark Speedups for Sequent Symmetry(32.0) 8.060.0EmbarGridPoissonSparseEmbarGridPoissonSparse55.050.06.045.040.035.04.030.025.020.02.015.0(4.0)10.01.05.00.00102030Figure 2: Benchmark Speedups for Intel Paragon[21] D.
A. Reed, R. A. Aydt, T. M. Madhyastha, Roger J.Noe, Keith A. Shields, B. W. Schwartz, An Overviewof the Pablo Performance Analysis Environment. Department of Computer Science, University of Illinois,November 1992.[22] Z. Segall, et al., An Integrated Instrumentation Environment, IEEE Transactions on Computers, Vol. 32,No. 1, pp. 4{14, Jan., 1993.[23] Z. Segall and L. Rudolph, PIE: A Programming andInstrumentation Environment for Parallel Processiong,IEEE Software, Vol. 2, No. 6, pp. 22{37, Nov., 1985.[24] V.
S. Sunderam, PVM: A Framework for Parallel Distributed Computing, Concurrency: Practice & Experience, Vol. 2, No. 4, pp. 315{339, December 1990.0.00102030405060Figure 4: Benchmark Speedups for BBN TC200060.0EmbarGridPoissonSparse55.050.045.040.035.030.025.020.015.010.05.00.00102030405060Figure 5: Benchmark Speedups for KSR KSR-1setupsystem setup195.090.085.080.075.070.065.060.055.050.045.040.0(1)-(8)35.030.025.0(3)-(7)*20.015.010.05.00.0010 20fork2...main3elementallocationuser4user setupparallel5...parallelalgorithm6user wrapup7fft(5)-(6)cyclic(4)-(7)(3)-(7)(2)-(8)304050#nodes60708090system wrapup8Figure 8: Detailed Speedup Prole: Poisson on BBNTC2000Figure 6: pC++ Execution Phases for Measurement60.020.0fft55.015.0fft50.0(5)-(6)45.0(5)-(6)40.035.0cyclic10.0(3)-(7)*(4)-(7)(3)-(7)(2)-(8)5.0(1)-(8)30.0cyclic25.020.0(3)-(7)*(4)-(7)(3)-(7)(2)-(8)(1)-(8)15.010.05.00.0010#nodes20Figure 7: Detailed Speedup Prole: Poisson on Sequent Symmetry0.00102030#nodes405060Figure 9: Detailed Speedup Prole: Poisson on KSRKSR-1.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
