Subject:
RE: C516:An Analysis of VI Architecture Primitives in Support of Parallel and Distributed Communication
From:
"Philip Buonadonna" <philipb@eecs.berkeley.edu>
Date:
Fri, 7 Sep 2001 18:07:20 -0700
To:
"Geoffrey Fox" <fox@csit.fsu.edu>
CC:
mdw@cs.berkeley.edu
X-UIDL:
89daf901ef440200
X-Mozilla-Status:
0011
X-Mozilla-Status2:
00000000
Received:
by mailer.csit.fsu.edu (mbox gcfpc) (with Cubic Circle's cucipop (v1.31 1998/05/13) Sat Sep 15 19:31:10 2001)
X-From_:
fox@mailer.csit.fsu.edu Sat Sep 15 19:30:58 2001
Return-Path:
<fox@mailer.csit.fsu.edu>
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 327A023A03 for <gcfpc@csit.fsu.edu>; Sat, 15 Sep 2001 19:30:58 -0400 (EDT)
Received:
from localhost by dirac.csit.fsu.edu (AIX4.2/UCB 8.7) id TAA56174; Sat, 15 Sep 2001 19:30:57 -0400 (EDT)
Resent-Message-Id:
<200109152330.TAA56174@dirac.csit.fsu.edu>
Delivered-To:
fox@csit.fsu.edu
Received:
from ups.millennium.berkeley.edu (ups.Millennium.Berkeley.EDU [169.229.48.61]) by mailer.csit.fsu.edu (Postfix) with ESMTP id 5EA5223A0B for <fox@csit.fsu.edu>; Fri, 7 Sep 2001 21:07:24 -0400 (EDT)
content-class:
urn:content-classes:message
MIME-Version:
1.0
Content-Type:
multipart/mixed; boundary="----_=_NextPart_001_01C13802.9C465436"
X-MimeOLE:
Produced By Microsoft Exchange V6.0.4417.0
Message-ID:
<B7D7A2ED64049746BBCDFB0DA028CC54061DFB@ups.millennium.berkeley.edu>
Thread-Topic:
C516:An Analysis of VI Architecture Primitives in Support of Parallel and Distributed Communication
Thread-Index:
AcEo7OQoI/OYzo0bTQi/UrzkVso2CwO/s37A
Resent-To:
Geoffrey Fox <gcfpc@csit.fsu.edu>
Resent-Date:
Sat, 15 Sep 2001 19:30:56 -0400
Resent-From:
Geoffrey Fox <fox@mailer.csit.fsu.edu>


Thanks for taking the time to review our paper. We have completed
revisions and are submitting the final for publication.  

We have addressed many of the issues cited by the reviewers. Per referee
comments are below:

Referee 1 comments:

A study of AM and split-c over software based implementations might be
interesting, but would require additional time in terms of setting up
and benchmarking. As well, MVIA only runs on certain adaptors that would
have to be acquired for the tests.  Porting TCP codes to VIA is a
separate work that is already being undertaken. We assumed a reliable
transport in order to measure the mapping cost.  Overall performance
effects on message loss would require a separate study.

Referee 3 comments:

While the implementation is important, high levels of detail would
quickly form the bulk of the paper and detract from the results. We
believe that keeping these sections brief and only presenting pertinent
details brings more focus to the analysis we are trying to perform.  

We decided to retain figures 3 & 4 as they give a visualization of where
the bulk of the overhead in the mapping is.

Referee 4 comments:

One of the key points of the paper is the impact of VI-based
communcation, of which Infiniband is a part.  We are trying to show what
the impact of VI/Infiniband-like communication primitives has on host
performance. Thus, a discussion of Infiniband is warranted.  

With respect to the insertion of additional figures, the individual
discussions of VIA, Infiniband, AM and split-c are presented for
completeness.  We have inserted references into these sections to make
it easier for the interested reader to obtain additional information.