Academic Commons


A Note on Implementing OPS5 Production Systems on DADO

Stolfo, Salvatore

This brief note is written in response to a recent publication, "Implementing OPS5 Production Systems on DADO," published as a Carnegie-Mellon Department of Computer Science Technical Report by Anoop Gupta in March, 1984. Gupta's paper analyzes the performance of OPS5 Production System programs on the DADO parallel computer, a special purpose production system (PS) machine. The analysis leads Gupta to conclude that DADO is not an effective OPS5 PS machine. we have studied Gupta's analysis carefully, and conclude that his conclusions are Inaccurate, flawed by, at times, incorrect or outdated information about the DADO2 prototype, and, at other times, inexact reasoning. We have divided the following into two sections. Section 2 details specific technical errors regarding the DADO2 design cited by Gupta. Although Gupta properly cites statistics we reported in earlier papers proposing DADO, his analysis is based on an earlier design of the second prototype system presently near completion at Columbia University. However, after making the changes appropriate to be consistent with the current technical design of DADO2, Gupta's analysis is also flawed in not adequately understanding the detailed workings of several reported algorithms. Section 3 focuses on philosophical differences. We shall be careful to accurately quote Gupta to strengthen our case that his conclusion is rather weak. We conclude that DADO indeed is an effective OPS5 processor. More importantly, we believe the DADO machine will produce dramatic performance Improvements of AI computation when the sequentialities inherent in OPSS are removed. The reader should first carefully read Gupta's paper and anyone of the most recent reports detailing the DADO system and algorithms, (Stolfo and Miranker, 1984), for example.



More About This Work

Academic Units
Computer Science
Department of Computer Science, Columbia University
Columbia University Computer Science Technical Reports, CUCS-130-84
Published Here
February 22, 2012