Pancake, C. M. and C. Cook, "What Users Need in Parallel Tool Support: Survey Results and Analysis", to appear in The 1994 Scalable High Performance Computing Conference, May 1994, Knoxville, TN.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....predict, and tune the performance of their parallel and distributed applications using broad range of approaches. Most of the developed tools tend to be used at runtime, using performance data gathered by instrumenting the application program [2,3,4,5,8] A survey conducted by Pancake and Cook [6] has revealed the fact that tool use is still appallingly low among the high performance computing community. Cherri et al. 7] have cited the following three critical causes for this situation; 1) current tools are difficult to understand by scientific users, 2) tools vary widely across ....

C.M. Pancake and C. Cook, "What users Need in Parallel Tool Support: Survey Results and Analysis," Proceeding of Scalable High-Performance Computing Conference, Knoxville, Tennessee, May 1994.

....not be an understatement to say that parallel software is in a state of crisis. The vast majority of scientific programmers find that current parallel software support is inadequate [53] In fact, they are more likely to develop their own in house software support rather than use existing products [113]. The most commonly used parallel programming paradigm today is message passing. Standardization efforts have resulted in a portable message passing library called MPI (Message 1 2 Passing Interface) 105] Unfortunately, programming with message passing is tedious, as the programmer must ....

C. M. Pancake and C. Cook, What users need in parallel tool support: Survey results and analysis, in Proceedings of the 1994 Scalable High Performance Computing Conference, May 1994.

....architectures like multiprocessor and multicomputer systems. However, the development and maintenance of software for parallel computers requires signi cantly higher e orts than for sequential ones. A major part of this extra time and e ort is spent on testing, debugging, and optimizing the code [23]. Performing these tasks eciently requires elaborate tool support. Among the tools used, on line tools provide the most bene ts. On line tools in con Partly funded by the German Science Foundation, Contract SFB 342, TP A1 1 trast to o line tools are applied to an application while it ....

C. M. Pancake and C. Cook. What users need in parallel tool support: Survey results and analysis. In IEEE, editor, Proceedings of the Scalable High-Performance Computing Conference, May 23-25, 1994, Knoxville, Tennessee, pages 40-47, 1109 Spring Street, Suite 300, Silver Spring, MD 20910, USA, 1994. IEEE Computer Society Press.

....and so on often lead to dramatic changes in performance, sometimes in the wrong direction . Rerunning the same program with the same data often produces significantly difference performance results. Despite these challenges, many parallel computer users and vendors doubt the value of tools [8]. Users try to debug their programs with print statements, and they avoid runtime software monitoring of programs because of its overhead or side effects. Early high performance computers came with almost no tools, and programmers often resorted to assembler language. For example, originally ....

Pancake, C., and Cook, C. What users need in parallel tool support: Survey results and analysis. In Proceedings of Scalable High-Performance Computing Conference (Knoxville, TN) (May 23--25 1994), IEEE Computer Society, pp. 40--47. http:/www.cs.orst.edu/ pancake/surveys/surveys.html. 20

....2.3 Implementation Architectures Many of the choices made in the development of the characteristics of the interface in this work involved consideration of what developers of the applications in the targeted fields would need and like. A study of these issues is presented by Pancake et al. 18] [19]; while their work was centered on scientific problems, with a few exceptions their observations and conclusions are equally applicable to other problem domains. We borrow from their discussion, as we consider the characteristics desirable in VLSI CAD and similar domains. 17 2.3.1 Algorithm ....

C. M. Pancake and C. Cook, "What users need in parallel tool support: Survey results and analysis," in Proceedings of the Scalable High-Performance Computing Conference, pp. 40--47, May 1994.

....HPC hardware and software technology for developing computational science applications, but scientists often lack sufficient background in parallel processing to understand and apply that technology. Experience has shown that it is difficult to make programming and analysis tools easy to use [5]. Generally, difficulties arise when the scientist s conceptual and operational models of the application are not well supported by the tool. A simple example was seen in early versions of ParaGraph [15] where displays of communication behavior clearly showed performance bottlenecks, but users ....

C. Cook and C. Pancake, What Users Need in Parallel Tool Support: Survey Results and Analysis, in Proc. of the Scalable High Performance Computing Conference, I. C. S. Press, ed., 1994, pp. 40--47.

....topology. 3. Primary Design Goals Tools that are used to monitor the run time behavior of parallel programs, like debuggers, must meet a couple of requirements in order to be of practical use. To meet these requirements has been a primary design goal for DETOP. Ease of use. Several surveys (e.g. [9]) show that one reason for tools not being widely used is the lack of user friendliness. Usage of tools should be clear and intuitive. Graphical interfaces and strictly guided user interaction can simplify the handling for unexperienced users. However, the interface must also provide a quick (and ....

C. Pancake and C. Cook. What Users Need in Parallel Tool Support: Survey Results and Analysis. Technical Report CSTR 94-80-3, Oregon State University, Dept. of Computer Science, Corvallis, OR, June 1994.

....settings, run the modelling program against each set of files, collate and merge the output files and finally produce some form of condensed output. For example, the output may simply be a low dimensional plot (often two or three dimensions) against the input parameters. For any one run, Pancake [9] has measured that users spend in the order of 10 of time performing job setup; for a multiple scenario experiment this may consume even more time because the operation of parametisation and control are more complex. If the job set up is performed manually, great care must be taken to ensure that ....

Pancake, C., Cook, C., "What Users Need in Parallel Tool Support: Survey Results and Analysis", Proceedings of 1994 Scalable High-Performance Computing Conference, May 23-25, Knoxville, Tennessee.

....One approach to facilitating the growth of parallel and distributed programming involves the development of new programming languages and new hardware. Recently, however, researchers have begun to focus on the importance of good software tools to assist developers of concurrent programs[Che93, PC94] These tools include debuggers, performance monitors, execution analysis and replay tools, and other aids. Our research also has focused on tools for program understanding and development, but we have a different emphasis: software visualization tools[SP92] A key component of understanding a ....

Cherri M. Pancake and Cutis Cook. What users need in parallel tool support: Survey results and analysis. In Proceedings of the Scalable High Performance Computing Conference (SHPCC '94), pages 40--47, Knoxville, TN, May 1994.

....provides insight into how to develop a methodology of performance debugging, it is not sufficient. Much to the dismay of tool builders, it has been observed that even on 16 platforms with fairly sophisticated performance tools, most parallel programers do not use the available performance tools[72]. Instead, programmers often manually insert code into their program to understand its performance. Looking at the way in which programmers manually search for bottlenecks provides hints about how to automate the process. For example, a common way to look at programs performance is to insert calls ....

C. M. Pancake and C. Cook, "What Users Need in Parallel Tools Support: Survey Results and Analysis", 1994 Scalable High-Performance Computing Conference, May 1994, pp. 40-47.

....axes. The W 3 model gives a well structured picture of the axes along which performance bottlenecks have to be tracked down. It lacks, though, on a general concept on how to do the refinement along these axes in a series of small steps. Clear and intuitive user interface Several surveys (e.g. PC94] show that a main reason for tools not being widely used is the lack of userfriendliness. Usage of tools should be clear and intuitive. Graphical interfaces and strongly guided user interaction can simplify the handling for unexperienced users. However, the interface must also provide a quick ....

C. Pancake and C. Cook. What Users Need in Parallel Tool Support: Survey Results and Analysis. Technical Report CSTR 94-80-3, Oregon State University, Dept. of Computer Science, Corvallis, OR, June 1994.

....scientists, with the remainder dominated by scientists and engineers. This paper addresses 2 the responses of the 326 vector and serial users, in some cases comparing those responses to those of the 448 parallel users. A more detailed analysis of the parallel group has been reported elsewhere [8]. For clarity and to facilitate comparison, results from the survey instrument s nine questions have been grouped into three areas for discussion. The first establishes a general characterization of the subject population. This is followed by the results for items dealing with respondents ....

....who currently under utilize tools, may be likely to underestimate their importance for parallel programming. 6. Conclusions The survey results indicate several disturbing trends in user expectations concerning parallel computing. In comparison with our studies of parallel programmers (e.g. [7,8]) the respondents appear to be overly optimistic about the ease of learning parallel programming as well as the potential for performance improvement. If past experiences are any indicator of future problems, parallel computing runs the risk of serious negative publicity when and if these users ....

[Article contains additional citation context not shown here]

Pancake, C. M. and C. Cook, "What Users Need in Parallel Tool Support: Survey Results and Analysis", to appear in The 1994 Scalable High Performance Computing Conference, May 1994, Knoxville, TN.

....scientists, with the remainder dominated by scientists and engineers. This paper addresses 3 the responses of the 326 vector and serial users, in some cases comparing those responses to those of the 448 parallel users. A more detailed analysis of the parallel group has been reported elsewhere [8]. For clarity and to facilitate comparison, results from the survey instrument s nine questions have been grouped into three areas for discussion. The first establishes a general characterization of the subject population. This is followed by the results for items dealing with respondents ....

....users, who currently under utilize tools, may be likely to underestimate their importance for parallel programming. 6. Conclusions The survey results indicate several disturbing trends in user expectations concerning parallel computing. In comparison with our studies of parallel programmers (e.g. [7,8]) the respondents appear to be overly optimistic about the ease of learning parallel programming as well as the potential for performance improvement. If past experiences are any indicator of future problems, parallel computing runs the risk of serious negative publicity when and if these users ....

[Article contains additional citation context not shown here]

Pancake, C. M. and C. Cook, "What Users Need in Parallel Tool Support: Survey Results and Analysis", to appear in The 1994 Scalable High Performance Computing Conference, May 1994, Knoxville, TN.

No context found.

Pancake, C., Cook, C., "What Users Need in Parallel Tool Support: Survey Results and Analysis", Proceedings of 1994 Scalable HighPerformance Computing Conference, May 23-25, Knoxville, Tennessee, 1994.

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