Alpern, B. and L. Carter (1995, February). The Myth of Scalable High Performance. In SIAM Conference on Parallel Processing for Scientific Computing.  Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Performance Evaluation Of The Parallel Fast Multipole.. - Ioana Banicescu (Correct)
....large initial problem size is used. For example, it is possible that reasonable speedups can be obtained by simply adding processors without increasing problem size. Thus, scaled speedup is not effective at identifying an invariant of application performance such as the sequential fraction (Alpern and Carter 1995). Gupta and Kumar have suggested that the isoefficiency metric be used for scalability measurement of parallel algorithms (Gupta and Kumar 1993) This metric assumes that a relationship between work and processors exists at a constant parallel efficiency and that it identifies this relationship. ....
.... Among the deficiencies of the isoefficiency metric are the following: i) it is not able to identify the number of processors required before an algorithm becomes an effective option, ii) it discounts otherwise valuable parallel algorithms for which an isoefficiency function does not exist (Alpern and Carter 1995). The isoefficiency metric can identify whether algorithms are scalable, but provides no insight as to the specific conditions required to establish an algorithm s superiority over competing algorithms in terms of scalability and it does not consider cost. It has been suggested that any relevant ....
Alpern, B. and L. Carter (1995, February). The Myth of Scalable High Performance. In SIAM Conference on Parallel Processing for Scientific Computing.
The Optimal Effectiveness Metric for Parallel Application Analysis - Edward Luke (Correct)
....curve can be made to look arbitrarily good by the selection of a sufficiently large initial problem size. One could argue that scaled speedup is not effective at identifying an invariant of application performance such as the sequential fraction. Arguments along these lines have been presented in [1]. Kumar has suggested the isoefficiency metric for scalability measurement of parallel algorithms. This metric identifies a function between work and processors that maintains a constant parallel efficiency. The basic idea behind this analysis is that the algorithm that can add processors at the ....
....perspective is that it does not identify the number of processors required before an algorithm becomes an effective option. In addition, it discounts otherwise valuable parallel algorithms for which an isoefficiency function does not exist. Alpern and Carter underscore these criticisms in [1]. One could summarize these arguments by stating that the isoefficiency metric is a qualitative metric rather than quantitative. That is, the isoefficiency metric can identify algorithms that are scalable but provides little insight as to the specific conditions required to establish an ....
B. Alpern and L. Carter. The Myth of Scalable High Performance. In SIAM Conference on Parallel Processing for Scientific Computing, February 1995.
The Optimal Effectiveness Metric for Parallel Application.. - Luke, Banicescu, Li (Correct)
....that reasonable speedups can be obtained by simply adding processors without increasing problem size. One could argue that scaled speedup is not effective at identifying an invariant of application performance such as the sequential fraction. Arguments along these lines have been presented in [1] and [16] Kumar has suggested the isoefficiency metric for scalability measurement of parallel algorithms. This metric identifies the relationship between work and processors that maintains a constant parallel efficiency, assuming such a relationship exists. The basic idea behind this analysis is ....
....perspective is that it does not identify the number of processors required before an algorithm becomes an effective option. In addition, it discounts otherwise valuable parallel algorithms for which an isoefficiency function does not exist. Alpern and Carter underscore these criticisms in [1]. One could summarize these arguments by stating that the isoefficiency metric is a qualitative metric rather than quantitative. That is, the isoefficiency metric can identify when algorithms are scalable but provides little insight as to the specific conditions required to establish an ....
B. Alpern and L. Carter. The Myth of Scalable High Performance. In SIAM Conference on Par8 allel Processing for Scientific Computing, February 1995.
Bibliography on the Solution of Sparse Linear Systems and Related .. - Arantes (1997) (Correct)
No context found.
Alpern, B. and Carter, L. The myth of scalable high performance. Tech. Report, Computer Science and Eng. Dept., San Diego Supercomputer Center, Univ. of California, San Diego, 1995.
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