Subject: CP&E #C396 Review Resent-Date: Sat, 18 Sep 1999 11:10:15 -0400 Resent-From: Geoffrey Fox Resent-To: p_gcf@npac.syr.edu Date: Mon, 25 Jan 1999 14:30:35 -0600 From: Ian Foster To: gcf@npac.syr.edu CC: foster@antares.mcs.anl.gov Paper Number: C396 Paper Title: On the Implementation of the Opus Coordination Language Authors: E. Laure, M. Haines, P. Mehrotra, H. Zima Referee: Ian Foster, foster@mcs.anl.gov Recommendation: Publish as is Referee Comments: Opus is an interesting system, and a description of its implementation is certainly appropriate for publication in CP&E. This paper provides such a description, and while it is clearly somewhat preliminary work (as far as one can tell, only the most basic elements of the proposed implementation exist), I nevertheless think that it is worth publishing. A few miscellaneous comments on the paper follow. While I don't require that these be addressed for publication, the paper would be stronger if they were: * The lack of any meaningful performance data was disappointing. How about some data for a real application; some detailed analysis of the data you have provided; or a comparison with an alternative implementation? Your current data don't really tell me anything. * I found the arguments advanced to justify the development of a new runtime system rather weak: the simultaneous statements that "other runtime systems are too complex" and "we currently suffer from a lack of support for heterogeneous systems, etc., which other runtime systems offer" is indicative of the difficulties. It would be good to see deeper arguments, and/or more detailed comparisons. * The techniques used to implement interactions between distributed SDAs will presumably be very similar to those used within e.g. Fx or HPF/MPI for inter-task communication.