<!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<meta content="text/html;charset=windows-1252"
http-equiv="Content-Type">
</head>
<body bgcolor="#ffffff" text="#000000">
All,<br>
<br>
Please make sure that a grid is not a requirement, but rather an
implementation to meet a requirement. The requirements need to be
carefully laid out, all the way from calibration and reconstruction,
through the first stages of analysis, all the way to the last stages of
interactive and ad-hoc analyses.<br>
<br>
It is fine to state requirements for use of data by off-site university
researchers, but it is not fine to state that you need to have 50% of
the compute capacity off-site and therefore you need a grid. Please
also in your plan include network bandwidth requirements to achieve
your goals within your reference architecture, and a time line for
deployment and stress testing. It is fine to state intangible benefits
of a more distributed computing model, including the training of future
computer scientists, especially if your CS department is engaged.<br>
<br>
Robust support of a full LHC style grid will be a costly item in labor,
both at Jefferson Lab and at universities, and if that is the best
solution to Jefferson Lab's science needs, then a careful case will
have to be built to get those funds allocated to this task. While we
can get some expert advice from our HEP colleagues, HEP will not want
to fund a 24/7 grid operations center for NP, so operational support
will not be free. It is now time to start thinking about the required
uptime of all the various components in the 12 GeV analysis system, and
the implied labor for that. The LHC grid is now running well, but the
operational costs are not small.<br>
<br>
We certainly expect to be looking at this aspect of the 12 GeV
computing plans at the upcoming internal review.<br>
<br>
regards,<br>
Chip<br>
<br>
<br>
Curtis A. Meyer wrote:
<blockquote cite="mid:4D559BB7.1030904@ernest.phys.cmu.edu" type="cite">
<pre wrap="">Thanks David -
I concur with Richard that the grid infrastructure is critical and
needs to be placed
into the plan somewhere. Also, in going through the numbers in the plan,
I did not
fully understand all the SpecInt numbers that were there. It would be
useful to have a
meeting where we go through these numbers before submitting the final
plan. I am
sure that they are right, but I am just "slow".
Curtis
On 2/11/11 1:38 PM, David Lawrence wrote:
</pre>
<blockquote type="cite">
<pre wrap="">Hi All,
Just a follow up to the discussion at the meeting on Wednesday. I'd
mentioned that Mark had submitted the document I wrote up in 2008 when
filling out Chip's original spreadsheet. The document titled "12GeV
Computing Plan" is GlueX-doc-1431. As to the specific issue of storing
simulated data, it has this in the second paragraph on page 2:
"Hall-D currently plans to follow a computing model that stores very
little simulated data on tape. The expectation is that it would be
nearly as fast or faster to reproduce the data as it would be to read it
off of tape and reprocess it. The “DST” files produced by the simulation
are therefore only expected to need 10% of the disk space used for the
raw, simulated data that will exist only on the farm nodes."
This was not backed up by any hard numbers but I still suspect it is
true. This refers only to what is needed regarding JLab resources and
not what may be implemented outside of JLab (and JLab's budget). To my
knowledge this represents what is our current Computing Plan for Hall-D.
Regards,
-David
_______________________________________________
Halld-offline mailing list
<a class="moz-txt-link-abbreviated" href="mailto:Halld-offline@jlab.org">Halld-offline@jlab.org</a>
<a class="moz-txt-link-freetext" href="https://mailman.jlab.org/mailman/listinfo/halld-offline">https://mailman.jlab.org/mailman/listinfo/halld-offline</a>
</pre>
</blockquote>
<pre wrap=""><!---->
</pre>
</blockquote>
</body>
</html>