|
|
|
Data flow computer architecture Chudík J., Springer-Verlag, London, UK, 1984. Type: Book (9789780387136578) |
|
|
|
|
|
|
|
This chapter is a skillful summary of the data flow approach to computer system design. The chapter begins with a synopsis of the rationale for the data flow approach, followed by a discussion of differences between Data Flow (DF) and von Neumann or Control Flow (CF) models of computation. The discussion is illustrated by representing a data flow program as a directed graph. The concept of a token is introduced to represent the dynamics of DF computation. Next, the topic of data flow programming is introduced. This section is, perhaps, the most disappointing. Current debate concerning the proper approach to computer languages for DF processors is given little attention. No mention is made of the difficulties with recursion, for example. Chudík should nevertheless be credited for stating that “it seems to be possible to translate simply from existing high-level programming into a DF language.” Indeed it is. This writer was involved in a successful effort that revised a FORTRAN compiler to generate code for a DF machine. The next topic is a tutorial on the differences between Search Mode configurable computers (SM) and Interconnection Mode configurable computers (IM). The article makes the important point that “these two models are sufficiently general to derive various concrete systems by different implementations of control units, functional units, and communications systems.” The article then gives a good survey of five specific DF architectures. These are: (1) Dennis and Misunas (see, for example, [1]), (2) Rumbaugh [2], (3) Arvind and Gostelow (e.g., [3]), (4) Gurd and Treleaven (e.g., [4]), and (5) Lau (e.g., [5]). The following description of these concrete data flow architectures clarifies and extends the previous sections, delineating the problem of a computer architect and demonstrating that data flow is a viable solution to the given problem. The article is clearly written--more so than is the norm for topics of this kind. Indeed, it is the crisp writing that makes this a valuable summary of the current work on data flow architecture. It should be read by anyone who is curious about where data flow fits in the current scheme of things. In fact, with the importance of data flow as a method for wresting more speed out of the available electronic building blocks, this paper should be read by anyone involved in the computer design who is not already informed about data flow.
|
|
Reviewer:
G. T. Boswell |
Review #: CR109116 |
|
|
1) |
Dennis, J. B.; and Misunas, D. P.A computer architecture for highly parallel signal processing, in Proc. of the CAM annual conf., 1974, 402–409. |
|
2) |
Dennis, J. B.First version of a data-flow procedure language, in Programming Symposium (Paris, April 9–11, 1974), B. Robinet (Ed.), Springer-Verlag, NY, 1974, 362–376. See <CR> 17, 11 (Nov. 1976), Rev. 30,558. |
|
3) |
Arvind; and Gostelow, K. P.A new interpreter for data flow schemes and its implications for computer architecture, Tech. Report 72, Information and Computer Science Division, Univ. of California, Irvine, 1975. |
|
4) |
Gurd, J.; and Treleaven, P.A highly parallel computer architecture, Tech. Rep. Dept. of Computer Science, Univ. of Manchester, Manchester, UK, 1976. |
|
5) |
Comte, D.; Durrieu, G.; Gelly, O.; Plas, A.; and Syre, J. C.Systeme LAU: Rapport sur le langage evolue, TEAU 9/1, Contract SESORI 74167, Document 5/3059, CERT, Toulouse, 1976. |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
E-Mail
This
Printer-Friendly
|
|
|
|
|
|
|