From owner-chemistry: at :ccl.net Sat Oct  1 14:05:00 2005
From: "benne278 benne278]![umn.edu" <owner-chemistry ~~ server.ccl.net>
To: CCL
Subject: CCL: W:hardware for computational chemistry calculations
Message-Id: <-29451-051001134429-29306-hJ8diRtwDgdyHOokCHE0ow ~~ server.ccl.net>
X-Original-From: benne278 <benne278_-_umn.edu>
Content-Type: TEXT/plain; CHARSET=US-ASCII
Date: Sat, 01 Oct 2005 11:59:19 CDT
MIME-Version: 1.0


Sent to CCL by: benne278 [benne278===umn.edu]
On 30 Sep 2005, Perry E. Metzger perry===piermont.com wrote:

>In my comment, I'm referring to the more general case

I know.  That is why I called hyperthreading a "special case".  There are
also some drawbacks to hyperthreading but they are not particularly
relevant to this thread.


> So, the question becomes, does the difference in speed for your app
> between having enough memory to hold the whole scratch file in buffer
> cache increase speed sufficiently to justify the marginal $1000
> cost? That depends on how I/O bound you are.


I'm not trying to provide a full decision tree here.  In fact, I'm
deliberately trying to avoid that because it will lead to another tedious
thread, and the original poster didn't provide enough information anyway. 
If the OP's quantum calculations use 30 GB of scratch, that will likely
change the answer vs. if they are only using 2 GB of scatch.  

The simple question was if SCSI ever has an advantage, and the simple
answer is that yes, sometimes it does, even if it isn't often.  Therefore,
it should be considered, even if you ultimately end up deciding the money
is better spent on something else.

If I might make a suggestion for people who post performance questions to
the list, it might help if they offered up short (~ 60 min?) test jobs.  

I bet there are a few people on the list who would be willing to run a
short example job on their hardware for you, and this will give you an
answer that is probably a lot more helpful than somebody listing all the
many variables that interact to determine performance.  You might even find
out that changing the code you are using would be a bigger boost than
buying faster hardware.

-Eric