Mengjuei -
Okay, I looked at the apple G5 pmemd performance optimizations a bit, and
I
had no idea we were talking about this level of performance optimization.
This stuff is written to a very low hardware-specific level, and would be
a
nightmare to maintain from our perspective (imagine each hardware vendor
doing this level of optimization of our critical code and saying "here,
you
support it"). Also, though you have not seen pmemd 9 checkins that are
different, that is because I won't check stuff in until we are further
along
with critical stuff like amoeba. There are significant changes everywhere
in the 9 codebase, and adapting this stuff to 9 will probably be
nontrivial.
I would recommend we let apple support this level of patch themselves,
releasing it on their applications website or whatever; if there is
popular
demand from the users, they will do it to sell hardware. Dave, I attached
all the source Mengjuei sent me so you can take a look too, and voice an
opinion on this sort of thing as El Jaffe. Personally, I shudder to think
of the maintenance and test impact (and it's not that I'm a total weenie;
heck, I used to both write and teach assembler, but I didn't do it
targetting 1000's of users on every machine in the universe that's not
slug
slow, and a few that are ;-)).
Regards - Bob
----- Original Message -----
From: "Mengjuei Hsieh" <mengjueh.uci.edu>
To: "Robert Duke" <rduke.email.unc.edu>
Sent: Friday, July 01, 2005 4:46 PM
Subject: Re: amber-developers: Apple's feed back for amber.
> Dear Dr. Duke,
>
> For the access to a G5 cluster, you can contact Warner Yuen at
> wyuen.apple.com, he is the consultant in the research computing group,
> Apple
> computer.
>
> I already did the JAC benchmark on their demo cluster, it seems to be
> working very well. See
> http://apple.sysbio.info/~mjhsieh/archives/000473.html for detail JAC
> benchmark.
>
> I also attached the patches, they are the source codes and some config.h
> files.
>
> Best,
> --
> Mengjuei Hsieh, Luo group, Molecular Biology and Biochemistry,
University
> of
> California Irvine. Tel: 49562, Address: 3144 Natural Science I building,
> UCI, Irvine CA 92697-3900. Group Homepage: http://rayl0.bio.uci.edu/
>
>
>> From: Robert Duke <rduke.email.unc.edu>
>> Reply-To: "amber-developers.scripps.edu" <amber-developers.scripps.edu>
>> Date: Fri, 24 Jun 2005 07:24:44 -0400
>> To: <amber-developers.scripps.edu>
>> Subject: Re: amber-developers: Apple's feed back for amber.
>>
>> Mengjuei -
>> If you send this stuff to me, I will check it out and issue an
"official
>> patch", provided someone can provide me access to a G5 cluster
somewhere
>> (anyone game or making a G5 available?).
>> Thanks - Bob Duke
>
>
Received on Wed Apr 05 2006 - 23:49:54 PDT
