M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications," IEEE Transactions on Computers, vol. C-37, no. 9, pp. 1088-1098, September 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....sets for VLIW and vector machines enable the compiler to express instruction independence statically. Superscalars take a different approach and examine many instructions in the execution stream simultaneously, violating the sequential ordering when they determine it is safe to do so. Prior work [15, 16, 17, 18] demonstrated that ample instruction level parallelism (ILP) exists within applications, but the control dependencies that sequential fetch introduces constrains this ILP. Despite tremendous effort over decades of computer architecture research, we have yet to devise a processor that comes close ....

A. M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications," IEEE Transactions on Computers, vol. 37, no. 9, pp. 1088--1098, 1988.

....job. We use the mean response time in this paper since it is most often the measure of interest to users of such systems. The parallelism profile of a job, defined as the number of processors an application is capable of using at any point in time during its execution, was introduced by Kumar [11]. More generally, a speedup function, Gamma, specifies the rate at which work is completed as a function of the number of processors allocated to it. Since parallel programs can have a wide variety of execution characteristics in practice, we consider a number of different classifications of jobs ....

M. Kumar. Measuring parallelism in computation-intensive scientific/engineering applications. IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

....jobs to be released at arbitrary times. Some results consider only fully parallelizable jobs (or equivalently time sharing with one processor) In practice, however, parallel programs can have a wide variety of execution characteristics. The parallelism profile of a job was introduced by Kumar [11] and extended by Deng and Koutsoupias [6] using a DAG model to represent the data dependency within the job. Turek et al. 25] model this more simply with a speedup function, Gamma. This specifies the rate at which work is completed as a function of the number of processors allocated to it. They ....

M. Kumar. Measuring parallelism in computation-intensive scientific/engineering applications. IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

....jobs to be released at arbitrary times. Some results consider only fully parallelizable jobs (or equivalently time sharing with one processor) In practice, however, parallel programs can have a wide variety of execution characteristics. The parallelism profile of a job was introduced by Kumar [11] and extended by Deng and Koutsoupias [6] using a DAG model to represent the data dependency within the job. Turek et al. 25] model this more simply with a speedup function, Gamma. This specifies the rate at which work is completed as a function of the number of processors allocated to it. They ....

M. Kumar. Measuring parallelism in computation-intensive scientific/engineering applications. IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

....may be obtainable by a static analysis of the code and the input file [DI89, B 91, Sar89a] but these techniques often fail outside of toy examples. 6 In a feeble attempt to circumvent the problem, some papers suggest that the users themselves provide some estimates of job parameters [MEB91, Kum88, PD89, Sev89] Even ignoring the fact that this is too far removed from current computing practice, we have to contend with the problem that users will abuse the system by quoting fake values for the parameters. 6 In contrast, estimating the t parameters of the machine (called performance ....

M. Kumar. Measuring parallelism in computation-intensive scientific /engineering applications. IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

..... There are a limited number of processors able to work on a program. There is unlimited parallelism in a program or program fragment, i.e. an additional processor can always be used. Put another way, the program is unsaturable. Such implied large degrees of parallelism are demonstrated in [4, 19]. The branch probability is known. Next, the following terms are defined (also see Figure 3) R processors are available in toto for the execution of a program. p is the probability of the branch executing true. This is an independent variable. w is the constant amount of work to be ....

Kumar, M. Measuring Parallelism in Computation-Intensive Scientific/Engineering Applications. IEEE Transactions on Computers 37(9):1088-1098, September, 1988.

....For this model, a harmonic mean speedup of a factor of 40 was achieved on the integer benchmarks; unlimited execution resources were assumed. For an Oracle (i.e. a branch predictor that obtains 100 accuracy) a speedup of 158 was obtained. For a limit study on scientific benchmarks, see: [8]. Lam and Wilson demonstrated that large ILP exists in integer codes, but concluded that it was unlikely to be realized, particularly with high IPC, because of the machine limitations extant at the time. In particular, no commercial machine realized MF, and few realized CD, although such ....

M. Kumar, "Measuring Parallelism in Computation-Intensive Scientific/Engineering Applications," IEEE Transactions on Computers, vol. 37, no. 9, pp. 1088-1098, September 1988. of 17

....this was an early study which ignored the problems presented by conditional code. The benchmarks used are very small by today s standards, most less than 200 cards, and many do not even contain DO loops. 17 In a study of very large benchmarks using parallelism time profiles for programs [92], Kumar shows that the amount of parallelism varies widely during the course of execution. Both the ideal case of full knowledge of control and data dependencies, and the case where control and data knowledge is restricted, show approximately the same amounts of parallelism. Average parallelism ....

M. Kumar, Measuring Parallelism in Computation-Intensive Scientific/Engineering Applications, IEEE Transactions on Computers C-37(9), 1988, pp. 10881098.

....job. We use the mean response time in this paper since it is most often the measure of interest to users of such systems. The parallelism pro le of a job, de ned as the number of processors an application is capable of using at any point in time during its execution, was introduced by Kumar [11]. More generally, a speedup function, speci es the rate at which work is completed as a function of the number of processors allocated to it. Since parallel programs can have a wide variety of execution characteristics in practice, we consider a number of di erent classi cations of jobs ....

M. Kumar. Measuring parallelism in computation-intensive scientic/engineering applications. IEEE Transactions on Computers, 37(9):1088-1098, September 1988.

....they can adjust the number of processors allocated to a job during its execution [9] In general, we can use parallelism profiles to characterize parallel jobs. A parallelism profile is defined as the number of processors an application is capable of using at any point in time during its execution [15]. During execution, if the parallelism of an application varies with time, it is said to have multiple phases of parallelism. Note that although our job model assumes linear speedup within each phase of parallelism, jobs with multiple phases of parallelism will execute with sublinear speedup. We ....

M. Kumar, Measuring parallelism in computation-intensive scientific/engineering applications, IEEE Trans. Comput., 37 (1988), pp. 1088--1098.

....information transparent, performance tuning is extremely difficult. P 3 T at compile time computes a set of performance parameters each of which reflects a different performance aspect. In the following all P 3 T performance parameters are described. 4. 1 Work Distribution It is well known [8, 6, 44, 42, 30, 40, 13, 41, 34, 25] that the work distribution has a strong influence on the cost performance ratio of a parallel system. An uneven work distribution may lead to a significant reduction in a program s performance. Therefore, providing both programmer and parallelizing compiler with a work distribution parameter for ....

M. Kumar. Measuring parallelism in computation intensive scientific/engineering applications . IEEE Transactions on Computers, 37(9):1088--1098, 1988.

....sequentially, while the second reflects the fraction of time during which the algorithm can be executed concurrently by all the available processors. The execution profile is defined as the number of processors kept busy as a function of time, given a fixed number of available processors [Kumar88] The execution profile is similar to the parallelism profile. The difference is that in the parallelism profile the number of available processors is unbounded. Periods of homogeneous processor utilization are manifested in execution profiles. These periods, called phases, can usually be ....

....are available. 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 ....

[Article contains additional citation context not shown here]

Kumar, M., "Measuring parallelism in computation-intensive scientific/engineering applications," IEEE Transactions on Computers, vol. 37, no. 9, pp. 1088-1098, September 1988.

....[12, 5] there is very little hard evidence from real measurements on parallel machines. The little work that has been done concentrates on modeling single applications, e.g. showing how a specific algorithm leads to changes in the degree of parallelism in different phases of the computation [18, 24, 20]. There is practically no work relating to the mix of different jobs that are found on parallel machines. This makes it hard to study and compare operating system policies for scheduling and processor allocation. Computer Sciences Corporation, NASA Contract NAS 2 12961. This study makes an ....

M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications". IEEE Trans. Comput. 37(9), pp. 1088--1098, Sep 1988.

No context found.

M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications," IEEE Transactions on Computers, vol. C-37, no. 9, pp. 1088-1098, September 1988.

No context found.

M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications," IEEE Transactions on Computers, Vol. 37, No. 9, pp. 10881098, 1988.

No context found.

M. Kumar, "Measuring parallelism in computation-intensive scientific/engineering applications, " IEEE Transactions on Computers, Vol. 37, No. 9, pp. 1088- 1098, 1988.

No context found.

M. Kumar. "Measuring parallelism in computation-intensive scientific/engineering applications." IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

No context found.

M. Kumar. Measuring parallelism in computation intensive scientific/engineering applications . IEEE Transactions on Computers, 37(9):1088--1098, 1988. 15

No context found.

M. Kumar. Measuring parallelism in computation intensive scientific/engineering applications . IEEE Transactions on Computers, 37(9):1088--1098, 1988.

No context found.

M. Kumar. Measuring parallelism in computation-intensive scientific /engineering applications. IEEE Transactions on Computers, 37(9), 1088-- 1098, September 1988.

No context found.

M. Kumar. Measuring parallelism in computation intensive scientific/engineering applications . IEEE Transactions on Computers, 37(9):1088--1098, 1988.

No context found.

A. M. Kumar, "Measuring parallelism in computationintensive scientific/engineering applications," IEEE Transactions on Computers, vol. 37, no. 9, 1988.

No context found.

M. Kumar, "Measuring Parallelism in Computation-Intensive Science Engineering Applications," IEEE Trans. Computers, vol. 37, no. 9, pp. 5--40, Sept. 1988.

No context found.

M. Kumar. "Measuring parallelism in computation-intensive scientific/engineering applications." IEEE Transactions on Computers, 37(9):1088--1098, September 1988.

No context found.

Manoj Kumar, `Measuring parallelism in computation-intensive scientific/engineering applications', IEEE Trans. Computers, 37, (9), 1088--1098 (1988).

First 50 documents  Next 50

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