The author sketches several types of multiprocessor configurations, discussing the advantages and disadvantages of various systems for placement and sharing of memories, processors, switches, buses, etc. He defines a “cost index,” (hardware cost per node) × (average message time delay), for a multiprocessor network in use. Cioffi then derives formulas for predicting this cost index based on various assumptions about message traffic, processor speeds, and other parameters. Using these formulas, he compares seven systems, each a plausible example of a different type of configuration. In a final section, he discusses alternative types of communication mechanisms.
This paper was unsatisfying in many ways. Having experience with one well-documented multiprocessor system (see [1]), I found the author’s statements about its inherent advantages and disadvantages to be ill-founded, and I am left with little confidence in the corresponding statements about architectures with which I am not familiar. The author’s classification of message communication mechanisms does not seem to have been well thought out; he refers to “synchronous” communication (unbuffered) and “asynchronous” communication (buffered) as if the latter somehow avoided the need for any synchronization by the sender, and as if the former were somehow more “safe.” In spite of these objections, the discussion of alternative architectures may serve as a useful introductions, and the cost index formulas may have some theoretical significance.
It is regrettable that the paper was not edited for its use of the English language, which sometimes seriously hinders understanding of the text. Also, the reviewer would have been better able to use the list of references if they had been specifically cited in the text.