Pancake, C. M., "Software Support for Parallel Computing: Where Are We Headed?" Communications of the ACM, 34 (11): 52--64 (Nov 1991).  Home/Search   Document Not in Database   Summary   ACM   TOC   Related Articles   Check

....to support computing in this new environment. Unfortunately, the software infrastructure (e.g. programming languages, compilers, operating systems, and performance monitoring and prediction tools) of 25 available today still has not kept pace with the state of the art multiprocessing hardware [Pan91, SMS95]. In particular, the lack of useful, accurate facilities for measuring and analyzing program performance is particularly distressing, since performance is the raison dtre for parallelism. 1.1 Performance Analysis Systems Traditional performance analysis systems (e.g. gprof) that generate ....

C. M. Pancake, Software Support for Parallel Computing: Where Are We Headed? Communications of the ACM, Vol. 34, No. 11, 1991, pp. 52-64. of 25

....behavior by monitoring program execution on an architecture also faces this question. This is especially true for the presentation and analysis of performance data using visual and aural means. Visualization and auralization can present fundamental information about program execution (e.g. [5, 15, 19, 27, 29, 33]) While they are increasingly being deemed powerful tools for parallel program monitoring and debugging, they are typically and justifiably considered to be more of an art than a science. What a user sees and hears can depend on his or her perception of the graphics and sound employed in the ....

Pancake, C., "Software Support for Parallel Computing: Where Are We Headed?", Comm. ACM, Nov. 1991.

....new environment. Despite the past 15 years of research on system software (e.g. programming languages, compilers, operating systems, and performance monitoring and prediction tools) the software infrastructure available today has not kept pace with the state of the art multiprocessing hardware [Pan91]. In particular, the lack of useful, accurate facilities for measuring and analyzing program performance is particularly distressing, since performance is the raison d tre for parallelism. 1.1 Performance Visualization Systems Traditional performance analysis systems ( e.g. gprof ) that generate ....

C. M. Pancake, "Software Support for Parallel Computing: Where Are We Headed?" Communications of the ACM, Vol. 34, No. 11, 1991, pp. 52-64.

....as the C language or FORTRAN with additional library for inter processor communication. But, that approach is not suitable for parallel programming. First, parallelism is likely to be too coarse grained in such ways. In so many applications such as natural language processing[21] genome computing[15], algorithms are described naturally only in a level which has fine granularity. When we write parallel programs in the way using libraries, resulting codes often create too many tasks for system to execute them efficiently. For task creations, task communications, task context switches, we must ....

Pancake, C. M., Software Support for Parallel Computing: Where are We Headed?, Communications of the ACM, pp. 53--64, 1991.

....become, or about to become, obsolete. The only reasonable business decision under this circumstances is to write code for some existing robust programming language, which presently would mean a serial programming language. This lack of a robust language is part of the par8 allel software crisis [P]. Following a meeting of an industry advisory board panel [P] pessimism is expressed about reaching agreement on a robust parallel programming language, because of the following. While, there seems to be agreement that a sharedmemory ( PRAM like ) language will be ideal from the user end, present ....

....decision under this circumstances is to write code for some existing robust programming language, which presently would mean a serial programming language. This lack of a robust language is part of the par8 allel software crisis [P] Following a meeting of an industry advisory board panel [P], pessimism is expressed about reaching agreement on a robust parallel programming language, because of the following. While, there seems to be agreement that a sharedmemory ( PRAM like ) language will be ideal from the user end, present parallel machines require that parallel programs include ....

C.M. Pancake. Software support for parallel computing: where are we headed? CACM, 34,11: 53--64, 1991.

....low priority activity. Yet in high performance computing, one of the major areas in which parallel processing is currently being applied, performance is clearly the dominant issue. In response to this problem, many researchers have developed tools and methodologies for parallel programming [1]. Typical first generation tools, developed largely by hardware vendors, and based on proprietary hardware, are suitable for raw programming, but are not sophisticated enough for large scale project use. The most recent approaches aim to provide a complete tool based environment for computer aided ....

C M Pancake. Software support for parallel computing: Where are we headed? Comm. of the ACM, 34(11):53--64, 1991.

....programs in Fortran, and that most scientists and engineers program only in Fortran. The argument was that programs in this newer language would be, oh, so much better because of the work in vectorizing. The scientists counter and I am afraid that we are not hearing this argument well[26] is that those old, empirically validated programs are the purpose of programming. Calculating the wrong answer quickly is not any help. Programs are not the object of science, knowledge is. Those old, antiquated programs are well tested and jive with the empirical relations observed in the ....

....well tested and jive with the empirical relations observed in the real world. We, computer science, are forgetting the Hamming dictum[19] The purpose of programming is insight, not numbers. The language debate, if indeed it is a debate, just will not go away. But are we asking the right questions[26] If we continue with an attitude[26] that the world is waiting with bated breath for the next program or programming language we will not endure as a discipline. If we continue imbuing our students with this attitude, we will continue to see declining enrollments as the sciences and ....

[Article contains additional citation context not shown here]

C. M. Pancake. Software support for parallel computing: Where are we headed? CACM, 34(11):53--64, 1991.

....nine classes, semantic properties a#ecting the development of parallel programs are compared. From a consideration of the general function of programming languages in software engineering, we derive basic requirements on parallel languages. 1 Introduction Phrases like parallel software crisis [52] or software dilemma [7] are commonly used when scientists discuss today s situation of parallel programming. To some degree, this is up to programming languages which provide only poor support for parallel software development as some of their properties hinder advanced analysis and ....

....data parallel algorithms concern arrays, many languages provide means to specify an alignment between arrays for minimizing communication overhead. Languages with Implicit Parallelism. In general, neither users nor application programmers are interested in parallelism itself, but in performance [52]. Therefore it is very reasonable to build a language which allows using parallel resources but does not require to explicitly specify which parts of a program can be executed in parallel. Instead, the language should not enforce any unnecessary specification of an execution order for independent ....

C.M. Pancake. Software Support for Parallel Computing: Where Are We Headed? Communications of the ACM, 34(11):53--64, November 1991.

....algorithms during the software development, some support for debugging and testing can also be integrated into the tool set without having to change the representation. The use of graphics and visual environments during the process of parallel software development is demanded by many experts [17, 12]. 3 Nevertheless, it is often neglected that graphics and cute pictures are no panacea for all problems in computer science, especially when dealing with complex systems. A visual tool for parallel programming must also offer concepts, principles and methodologies to aid the user more than just ....

....simple observation that most parallel programs contain two logical levels of parallelism: 1) coarse grain parallelism and (2) fine grain parallelism. 2 [In a parallel world] we have many more dimensions that affect performance of algorithms than we were accustomed to in the sequential world. [17], p. 56 3 : the group [Industry Advisory Board (IAB) agreed that high performance computing users are having trouble with the current text oriented tools,and that effective graphical techniques could alleviate many of their problems. 17] p. 1.1 Coarse Grain Parallelism The overall ....

[Article contains additional citation context not shown here]

C. M. Pancake. Software Support for Parallel Computing: Where Are We Headed? Communications of the ACM, 34(11):53--64, November 1991.

....an nCUBE 3200. 4.2 Analysis Using Performance Prediction Results The performance of parallel applications is affected by many factors which need not be considered with sequential programs. Consequently, performance analysis for parallel applications much take more of a multidimensional approach [17]. The following examples illustrate the utility of this symbolic performance model in analyzing the impact of several effects on a scalable application. A major use of performance prediction results is in the area of performance debugging. Figure 9 illustrates the percentage of time spent in ....

C. M. Pancake. Software support for parallel computing: Where are we headed? Communications of the ACM, 34(11):53--64, November 1991.

....: 26 5 Conclusion 27 D6H 1 Rel 0. 9 June 28 1994 Visual Parallel Programming PACT 1 Introduction The demand for supporting the programming of parallel machines by visual tools has often been stated in the literature [Karp 87] Panc 91] Chan 91] Glin 90] Although there exists a vast number of different graphical tools for parallel programming, the acceptance of these tools have not been very high, and even for the simplest of these tools a highly sophisticated knowledge of parallel programming is needed for proper use. A ....

....proper use. A reason for the low acceptance of these tools might be found in the neglection of the basic problems of visual programming and the widespread opinion that a tool is simple and easy to use, just because it has a graphical interface. David Gerlernter described the situation as follows [Panc 91] The potential power of graphic presentation of abstract data is tremendous, but [parallel] computer people are rank amateurs at this. The interface designs we ve seen so far are not nearly as informative as they could be. There is a tremendous body of knowledge and expertise on abstract ....

[Article contains additional citation context not shown here]

C. M. Pancake. "Software Support for Parallel Computing: Where Are We Headed?". Communications of the ACM, Vol. 34, No. 11, pp. 53--64, November 1991.

.... the exploitation of parallelism, have been developed in recent years; these include single assignment languages, such as Sisal [69] several functional languages [207] and process based languages, such as occam [111] However, the scientific user community is unwilling to accept new languages [172, 173], although it has been argued that the latter can satisfy the performance requirements of the users [34] Thus, most of today s parallel programming relies on enhanced versions of conventional sequential languages, such as Fortran, which dominates the field, C [98, 175] or C [212] Two basic ....

C. M. Pancake, "Software Support for Parallel Computing: Where Are We Headed?", Communications of the ACM, 34-11, Nov. 1991, pp. 52--64.

....We describe the language and briefly outline implementation issues of our prototype system. Keywords: concurrent logic language, programmed backtracking, transactions. 1 Introduction Advances in networking and parallel hardware are moving software development from programming to coordination [7, 27, 35]. Present computer systems can access a variety of remote information processing services, services that would be costly to implement and maintain locally, if at all possible. We would like to use these services not only in isolation, but to combine and coordinate them. As different machines offer ....

Pancake, C. Software Support for Parallel Computing: Where are we Headed? Commun. ACM , 34(11): 52-64, 1991.

....low priority activity. Yet in high performance computing, one of the major areas in which parallel processing is currently being applied, performance is clearly the dominant issue. In response to this problem, many researchers have developed tools and methodologies for parallel programming [1]. Typical first generation tools, developed largely by hardware vendors, and based on proprietary hardware, are suitable for raw programming, but are not sophisticated enough for large scale project use. The most recent approaches aim to provide a complete tool based environment for computer aided ....

C M Pancake. Software support for parallel computing: Where are we headed? Comm. of the ACM, 34(11):53--64, 1991.

....want is performance. It is the fact that going to parallelism is the only way to continue to enhance performance that makes parallelism a necessity [9: 90] The anticipated benefits have not always materialized, however, and they are associated with a high price tag in terms of programmer effort [6,3]. Is this another case of raising user expectations beyond the real capabilities of the underlying technology Our research addresses that issue in the larger context of user requirements for parallel software. In this paper, we analyze survey responses from potential parallel programmers ....

Pancake, C. M., "Software Support for Parallel Computing: Where Are We Headed?" Communications of the ACM, 34 (11): 52--64 (Nov 1991).

....want is performance. It is the fact that going to parallelism is the only way to continue to enhance performance that makes parallelism a necessity [9: 90] The anticipated benefits have not always materialized, however, and they are associated with a high price tag in terms of programmer effort [6,3]. Is this another case of raising user expectations beyond the real capabilities of the underlying technology Our research addresses that issue in the larger context of user requirements for parallel software. In this paper, we analyze survey responses from potential parallel programmers ....

Pancake, C. M., "Software Support for Parallel Computing: Where Are We Headed?" Communications of the ACM, 34 (11): 52--64 (Nov 1991).

No context found.

C. M. Pancake. Software support for parallel computing: Where are we headed? C. A. C. M., 34(11):53--64, 1991.

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