R. B. Irvin and B. P. Miller. A Performance Tool for High-Level Parallel Programming Languages. In Proceedings of the IFIP WG 10.3 Working Conference on Programming Environments for Massively Parallel Distributed Systems, pages  Home/Search   Document Not in Database   Summary   Related Articles   Check

....process both at the level of program events that need be traced and at the level of code regions where the tracing process must take place. The tool collects traces concerning high level source code events rather than low level run time environment events, thus no additional mapping is required [12] to present the user with the information collected in the traces. Therefore, ArrayTracer allows the user to stir the tracing process and focus it on selected parts of the application. Moreover, it facilitates the analysis of the application s behavior in terms of its internal structure (source ....

R. B. Irvin and B. P. Miller. "A Performance Tool for High-Level Parallel Programming Languages". Technical Report 1204, University of Madison-Wisconsin, 1993.

....both at the level of program events which should be traced and at the level of code regions where the tracing process should take place. The tool collects traces concerning high level source code events rather than low level run time environment s events, so no additional mapping is required [11] to present the information included in the traces to the user. Therefore, ArrayTracer allows the user to orientate the tracing process and focus it on selected parts of the application. Moreover, it facilitates the analysis of the application s behavior in terms of its internal structure (source ....

R. B. Irvin and B. P. Miller. "A Performance Tool for High-Level Parallel Programming Languages". Technical Report 1204, University of Madison-Wisconsin, 1993.

....data distributions. The 1 Templates and virtual processors provide an abstraction level which shields the programmer from the physical attributes of the target parallel computer. 14 Van Dongen et al. approach taken by the EPPP performance debugger belongs to a class of tools (such as ParaMap [12]) which are targeted for a specific high level model or language. By the way, ParaMap is designed for performance analysis of CM FORTRAN programs which are based on a significantly different parallel programming model than ours) The feedback provided by the EPPP performance debugger is ....

R. B. Irvin and B. P. Miller. A performance tool for high-level parallel programming languages. (grilled.cs.wisc.edu:technical papers/nv.ps.Z).

....to achieve higher performance. In this paper, we suggest an alternative to the existing approaches for performance debugging of data parallel programming based on user specified data distributions. The approach taken by the EPPP performance debugger belongs to a class of tools (such as ParaMap [6]) which are targeted for a specific high level model or language. By the way, ParaMap is designed for performance analysis of CM FORTRAN programs which are based on a significantly different parallel programming model than ours) The feedback provided by the EPPP performance debugger is ....

R. B. Irvin and B. P. Miller. A performance tool for high-level parallel programming languages. (grilled.cs.wisc.edu:technical papers/nv.ps.Z).

....this type arise in areas as diverse as computing, communications, manufacturing and transport. However, most of the work has concentrated on models involving a single job queue served by one or more processors. Very few results are available for systems with more than one queue. In a recent paper, [1], Mitrani and Wright examined a system with N parallel queues where the consequences of a breakdown are (a) the loss of all jobs in the corresponding queue and (b) the re direction or loss of all arrivals to that queue during the subsequent repair period. Those assumptions imply that the queue of ....

....at least in principle, for arbitrary N . Problems of comparison and optimization of routing policies can thus be tackled numerically. Several such numerical evaluations are reported. Various generalisations of the model, where the same solution methodology applies, are examined. References [1] I. Mitrani and P.E. Wright Routing in the Presence of Breakdowns, Performance Evaluation, 20, pp. 151 164, 1994. 2] N. Thomas and I. Mitrani Routing Among Different Nodes Where Servers Break Down Without Losing Jobs, Procs. IPDS 95, Erlangen, 1995. An Approach to Parallel Task Allocation ....

[Article contains additional citation context not shown here]

R. Bruce Irvin, Barton P. Miller, "A Performance Tool for High-Level Parallel Programming Languages", Technical Report Uni. Madison-Wisconsin, tr1204, 1993.

....for a module, procedure, loop, or statement. The complementary data centric view associates metrics with data structures and often provides new insights into a program s interaction with the memory system. Datacentric presentations have been used for distributed arrays in data parallel languages [14,15] and in cache tools for sequential programs [8,19] The data views that we added to Paradyn combine fine grained profiling (down to individual cache blocks) scalability (large data structures and large programs) and low overhead. To present this data, we added a new Paradyn resource hierarchy ....

R.B. Irvin and B.P. Miller, "A Performance Tool for HighLevel Parallel Programming Languages" in Programming Environments for Massively Parallel Distributed Systems, Birkaeuser Verlag, Basel, K.M. Decker and R.M. Rehmann, eds., 1994.

....objects. Data parallel systems must relate communication events to calls to the parallel libraries and to the data they operate on. The Paradyn system uses instrumentation of the binary to retrieve this information, and a tool NV to present the performance data at the desired level of abstraction [15]. pC knows specialized object types for data parallel arrays, and their tool uses this data function mapping to achieve the same goal [20] There is a correspondence between data parallel languages and Orca. Both use language level constructs (hidden library calls in dataparallel languages, ....

R.B. Irvin and B.P. Miller. A performance tool for highlevel parallel programming languages. In IFIP WG10.3 Working Conference on Programming Environments for Massively Parallel Distributed Systems, Ascona Switzerland, April 1994.

....communication are explicit (e.g. in message passing codes) For high level parallel languages, such tools can only capture and present dynamic performance data in terms of primitive operations (e.g. communication library calls) in the compilergenerated code. A few tools (Prism [13] NV [4], and MPP Apprentice [14] provide source level support for performance analysis of high level parallel languages. However, none of these tools provide source level performance support for the combination of data parallel languages and optimizing compilers necessary for Fortran D or HPF. ....

....users. 5 Related Work Few tools for dynamic performance analysis of parallel programs support abstract parallel languages, and few compilers export the requisite information needed to correlate dynamic performance data with the program source code. Notable exceptions include Prism [13] and NV [4] for CM Fortran, Forge90 [1] for Fortran 90 and HPF, and the MPP Apprentice performance tool, which supports C, Fortran and Fortran 90 on the Cray T3D [14] Our system differs from these tools in providing comprehensive support for data parallel programs by mapping runtime costs to data objects as ....

Irvin, R. B., and Miller, B. P. A Performance Tool for High-Level Parallel Programming Languages. In Programming Environments for Massively Parallel Distributed Systems (Basel, Switzerland, 1994), Birkhauser Verlag.

....the number of particles simulated, and other factors. The code was written by a group of computational scientists at the Numerical Aerodynamic Simulation Laboratory of NASA Ames Research Center. We briefly summarize our experience with PSICM; a full description of this study can be found elsewhere [15]. Our analysis of PSICM showed that combinations of code and data views of performance could be very useful for the performance analysis of data parallel programs. In particular, we identified a single statement that caused large amounts of implicit Node level activity and explained the cause of ....

R. Bruce Irvin. Performance tool for high-level parallel programming languages. Ph.D. Dissertation, University of Wisconsin -Madison, November 1995.

....real parallel applications. We conclude in Section 5. 2 THE NV MODEL OF HIGH LEVEL PERFORMANCE DATA We have developed a framework within which we can discuss performance characteristics of programs written in many different programming models. This framework is called the Noun Verb (NV) model [14]. In NV, a noun is any program element for which performance measurements can be made, and a verb is any potential action that might be taken by a noun or performed on a noun. The collection of nouns and verbs of a particular software or hardware layer defines a level of abstraction. Nouns and ....

R. Bruce Irvin and Barton P. Miller. A performance tool for high-level parallel programming languages. In Karsten M. Decker and Rene M. Rehmann, editors, Programming Environments for Massively Parallel Distributed Systems, pages 299--314. Birkhauser Verlag, 1994.

....the measured data equally across all sentences to which the measured sentence maps [1,60] However, such splitting assumes an equal distribution of low level work to high level code. It is often better to handle one to many mappings by merging the sentences to which the measured sentence maps [26]. The latter technique (used by the tool discussed in Chapter 5) makes no assumption about the distribution of performance data and helps to identify high level programming constructs whose implementations have been merged by an optimizing compiler. It also avoids misleading the programmer with ....

R. Bruce Irvin and Barton P. Miller. A performance tool for high-level parallel programming languages. In Karsten M. Decker and Rene M. Rehmann, editors, Programming Environments for Massively Parallel Distributed Systems, pages 299--314. Birkhauser Verlag, 1994.

No context found.

R. B. Irvin and B. P. Miller. A Performance Tool for High-Level Parallel Programming Languages. In Proceedings of the IFIP WG 10.3 Working Conference on Programming Environments for Massively Parallel Distributed Systems, pages

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