[Halld-offline] Fwd: 12 GeV computing requirements

[Mark M. Ito]

FYI.

-------- Original Message --------
Subject: 	12 GeV computing requirements
Date: 	Thu, 20 Jan 2011 09:30:05 -0500
From: 	Heyes Graham <heyes at jlab.org>
To: 	Weygand Dennis <weygand at jlab.org>
CC: 	Mark Ito M. <marki at jlab.org>, Stephen Wood A. <saw at jlab.org>, 
Robert Michaels <rom at jlab.org>, Brad Sawatzky <brads at jlab.org>



Here is the promised spreadsheet to try to capture computing requirements. There are probably other things that  I could have asked for but at least it's a start.

The orange colored cells of the spreadsheet are calculated using data from the uncolored ones so I would like you to fill in as much of the white part of the spreadsheet as you can.

In several places I ask for "time to process event", this needs a little bit of explaining. I am using our current generation 32-bit farm nodes as the standard. What processing time is required to process one event?

With a single threaded job running on a multi-core machine this time will be constant and equal to the CPU time per event. The multiple computing cores allow you to run more jobs in parallel but any single job runs for a CPU time equal to the number of events in the data file multiplied by the CPU time per event.

With the multi-threaded parallel architectures (CLARA and JENA) the CPU time per event decreases as the number of cores increases so the equivalent metric is the CPU time per event normalized by multiplying by the number of cores that part in the processing. If you can think of a better number feel free!

There are a couple of places where I ask for tape and disk estimates. Traditionally, based on experience with CLAS, we have assumed that the amount of work disk required is equal to 10% of the data taken in a year. With such large datasets in the 12 GeV era this simple minded calculation probably breaks down so feel free to estimate using a different method and let me know what that method is (educated guess is a good tool!). As far as tape estimate is concerned I assume that all the raw and all the analysis output will be stored but what about calibration and simulation?

As I wrote earlier this is a first try and may take a little (or a lot of) tweaking.

Thanks for your help,
	Graham



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110120/10cbd937/attachment-0002.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Offline Req 110120.xls
Type: application/octet-stream
Size: 24576 bytes
Desc: not available
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110120/10cbd937/attachment-0002.obj>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: Attached Message Part
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110120/10cbd937/attachment.ksh>