Subject: Paper C517: Automatic Determination of Matrix-Blocks From: "Achim Basermann" Date: Thu, 4 Oct 2001 16:25:16 +0000 To: gcf@indiana.edu CC: basermann@ccrl-nece.technopark.gmd.de, fox@csit.fsu.edu, fox@mailer.csit.fsu.edu X-UIDL: f9018b9ffe1f0000 X-Mozilla-Status: 0001 X-Mozilla-Status2: 00000000 Received: by mailer.csit.fsu.edu (mbox gcfpc) (with Cubic Circle's cucipop (v1.31 1998/05/13) Sun Oct 14 21:55:59 2001) X-From_: fox@mailer.csit.fsu.edu Sun Oct 14 21:54:10 2001 Return-Path: Delivered-To: gcfpc@csit.fsu.edu Received: from dirac.csit.fsu.edu (dirac.csit.fsu.edu [144.174.128.44]) by mailer.csit.fsu.edu (Postfix) with ESMTP id 83D0323A05 for ; Sun, 14 Oct 2001 21:54:10 -0400 (EDT) Received: from localhost by dirac.csit.fsu.edu (AIX4.2/UCB 8.7) id VAA92762; Sun, 14 Oct 2001 21:54:09 -0400 (EDT) Resent-Message-Id: <200110150154.VAA92762@dirac.csit.fsu.edu> Delivered-To: fox@csit.fsu.edu Received: from www.ccrl-nece.de (www.ccrl-nece.technopark.gmd.de [193.175.163.41]) by mailer.csit.fsu.edu (Postfix) with ESMTP id 7942523A23 for ; Thu, 4 Oct 2001 10:39:19 -0400 (EDT) Received: from ccrl-nece.de (gauss.ccrl-nece.technopark.gmd.de [193.175.160.57]) by www.ccrl-nece.de (8.9.3/8.9.3) with ESMTP id QAA00640; Thu, 4 Oct 2001 16:39:17 +0200 Received: from newton.ccrl-nece.technopark.gmd.de (sgi12 [193.175.160.117]) by ccrl-nece.de (8.9.3/8.8.5) with ESMTP id QAA03689; Thu, 4 Oct 2001 16:43:53 +0200 Received: by newton.ccrl-nece.technopark.gmd.de (980427.SGI.8.8.8) id QAA12392; Thu, 4 Oct 2001 16:25:17 +0200 (MESZ) Message-Id: <10110041625.ZM12387@newton.ccrl-nece.technopark.gmd.de> In-Reply-To: Geoffrey Fox "Request to review a paper C517: Automatic Determination of Matrix-Blocks" (Aug 19, 6:08pm) References: <3B804705.1050604@csit.fsu.edu> X-Mailer: Z-Mail (3.2.3 08feb96 MediaMail) Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Resent-To: Geoffrey Fox Resent-Date: Sun, 14 Oct 2001 21:54:09 -0400 Resent-From: Geoffrey Fox Dear Geoffrey, the referee report on the paper C517: Automatic Determination of Matrix-Blocks by Victor Eijkhout is included below. If you have question don't hesitate to send me a mail. With best regards, Achim CandC:PandE Referee Report Form *********************************************** Electronic Transimission to gcf@indiana.edu strongly preferred Referees Home Page: http://aspen.csit.fsu.edu/CandCPandE/ Email gcf@indiana.edu for URL of full paper to be reviewed WILEY Journal Home Page John Wiley and Sons, Ltd. Baffins Lane, Chichester West Sussex, PO19 1UD, England Telephone: (01243) 779777 Fax: (01243) 770379 REFEREE'S REPORT Concurrency and Computation:Practice and Experience ********** A: General Information Please return to: Geoffrey C. Fox Electronically Preferred gcf@indiana.edu Concurrency and Computation: Practice and Experience Computer Science Department 228 Lindley Hall Bloomington Indiana 47405 Office Phone 8128567977(Lab), 8128553788(CS) but best is cell phone 3152546387 FAX 8128567972 Please fill in Summary Conclusions (Sec. C) and details as appropriate in Secs. D, E and F. B: Refereeing Philosophy We encourage a broad range of readers and contributors. Please judge papers on their technical merit and separate comments on this from those on style and approach. Keep in mind the strong practical orientation that we are trying to give the journal. Note that the forms attached provide separate paper for comments that you wish only the editor to see and those that both the editor and author receive. Your identity will of course not be revealed to the author. C: Paper and Referee Metadata Paper Number Cnnn: C517 Date: October 4, 2001 Paper Title: Automatic Determination of Matrix-Blocks Author(s): Victor Eijkhout Referee: Achim Basermann Address: CCRL, NEC Europe Ltd., Rathausallee 10, D-53757 Sankt Augustin, Germany Referee Recommendations. Please indicate overall recommendations here, and details in following sections. publish as is accepted provided changes suggested are made reject Overall recommendations: accepted provided major changes suggested are made (almost reject) D: Referee Comments (For Editor Only) The manuscript describes two partitioning algorithms, one for regular and one for general matrices, to automatically determine natural block structures of the matrix. For parallel iterative solvers, this block structure can then be used for domain-decomposition based preconditioners. The manuscript is more a draft than a complete paper for a journal. The algorithms are only sketched, not explained in detail. Some figures would help in sections 2 and 3. In section 4, results are presented for a simple matrix struture only. Moreover, iteration counts are given for merely two simple, relatively small matrices. It is more interesting to see how the algorithms behave for large, irregularly structured matrices. It is also important to see how they can deal with many processors (domains), let's say more than 100. To prove the effectiveness and the efficiency of the algorithms described, a lot of test cases (also from different applications) have to be examined. An alternative approach is to use graph partitioning software like METIS. With the multi-objective option, also the values of the matrix non-zeros can be considered. From my experience, this gives at least as good iteration counts compared with even splitting and the Jacobi preconditioner as described for the partitioning algorithms in the paper. METIS definitely can handle unstructured cases. Thus also a comparison with METIS partitioning or other (graph) partitioning software would be helpful. E: Referee Comments (For Author and Editor) The manuscript describes two partitioning algorithms, one for regular and one for general matrices, to automatically determine natural block structures of the matrix. For parallel iterative solvers, this block structure can then be used for domain-decomposition based preconditioners. The algorithms are only sketched, not explained in detail. Some figures would help in sections 2 and 3. In section 4, results are presented for a simple matrix struture only. Moreover, iteration counts are given for merely two simple, relatively small matrices. It is more interesting to see how the algorithms behave for large, irregularly structured matrices. It is also important to see how they can deal with many processors (domains), let's say more than 100. To prove the effectiveness and the efficiency of the algorithms described, a lot of test cases (also from different applications) have to be examined. An alternative approach is to use graph partitioning software like METIS. With the multi-objective option, also the values of the matrix non-zeros can be considered. From my experience, this gives at least as good iteration counts compared with even splitting and the Jacobi preconditioner as described for the partitioning algorithms in the paper. METIS definitely can handle unstructured cases. Thus also a comparison with METIS partitioning or other (graph) partitioning software would be helpful. F: Presentation Changes Both the algorithmic and the result part have to be described in much more detail. For the algorithmic part, some figures would help a lot. In the result part, more test cases are necessary. A comparison with standard partitioning software would also be good. Furthermore, spelling has to be checked thoroughly. -- Dr.-Ing. Achim Basermann Technical Team Leader NEC Europe Ltd. C&C Research Laboratories Rathausallee 10 D-53757 Sankt Augustin Germany Tel.: +49 (0) 2241 - 92 52 - 95 Fax : +49 (0) 2241 - 92 52 - 99 e-mail: basermann@ccrl-nece.de www: http://www.ccrl-nece.de/~baserman/ .