[Halld-offline] computing model, alpha release

[David Lawrence]

Hi Mark,

     This is a great start. I have one quick question: have we verified 
that the reconstruction rate for the 10^8 data is the same as the 10^7? 
Even though the event rate to disk will be the same, I can imagine the 
10^8 data being busier and therefore, take longer to process. They are 
hopefully comparable, but we should verify it if it hasn't been already.

     I'd also like to follow up a little on the discussion started a 
while back on the how the GRID is to be included in the computing plan. 
It seems clear that our collaborators feel this will be an important 
part of the GlueX experiment's operation. Chip has indicated though that 
supporting it from the JLab side has significant resource costs and must 
be clearly justified. Has there been any discussion as to what exactly 
these costs are? I thought from earlier conversations that this was a 
manpower cost on the JLab side, but at this point, I don't know what is 
required. The Physics Division maintains several software packages 
already (ROOT, CERNLIB, Xerces,...) are there additional resources 
besides software maintenance? I know the our original CDR called for 
something like an OC24 connection due to the anticipated high bandwidth 
usage. Are there other hardware costs involved?

     We'll need to start gathering cost information as we develop the 
benefit argument in order to make a valid cost-benefit analysis.

Regards,
-David



On 2/15/11 2:33 PM, Heyes Graham wrote:
> Thanks Mark, that's number overload. I'll make a quick pass at 
> comparing this data with what Chip had from a few years ago.
>
> Maybe something by late tomorrow.
>
> Graham
>
> On Feb 15, 2011, at 2:24 pm, Mark M. Ito wrote:
>
>> Folks,
>>
>> Find attached a spreadsheet that estimates our computing needs. First 
>> the caveats:
>>
>>     * this is not the finished product
>>     * it has to be translated into terms for Graham's spreadsheet
>>     * some of the numbers are still guesses
>>     * PWA is nowhere included
>>     * there will have to be a GlueX Note on this eventually
>>     * things could be much better organized in the spreadsheet
>>     * my spreadsheet skill could use an upgrade
>>     * the estimates do not distinguish between JLab-resident and
>>       non-JLab-resident resources
>>     * the way we refer to "CPU's" needs to more sophisticated
>>       (SPEC-int-like).
>>
>> All of these will have to addressed. Still I thought that putting 
>> this out for discussion now would be better than waiting. So here it is.
>>
>> I am assuming steady state. The idea is that we need to do all of 
>> these things for each set of data we take (take the data, calibrate 
>> it, reconstruct it, compact it into multiple streams, simulate it, 
>> reconstruct the simulation...). If we do them all at a rate slower 
>> than that at which we take the data, we will fall farther and farther 
>> behind as the years go on. So to avoid that we have to keep up with 
>> the raw rate. There will be an initial ramp up and there may be long 
>> will be down periods, but eventually we will be doing all of these 
>> things on each set and I am not accounting in detail for the 
>> aforementioned edge effects.
>>
>> Note also that I assume the raw rate for this discussion is after a 
>> level three trigger, and that the at 10^7 the level three rejection 
>> rate is 0 and at 10^8 it is a factor of ten; i. e., independent of 
>> beam rate.
>>
>> On PWA: if people already have a good way of framing our needs in 
>> this area, let me know. I haven't looked through our notes or web 
>> pages on this yet.
>>
>>   -- Mark
>>
>>
>> <computing_model.xls>_______________________________________________
>> Halld-offline mailing list
>> Halld-offline at jlab.org <mailto:Halld-offline at jlab.org>
>> https://mailman.jlab.org/mailman/listinfo/halld-offline
>
>
> _______________________________________________
> Halld-offline mailing list
> Halld-offline at jlab.org
> https://mailman.jlab.org/mailman/listinfo/halld-offline
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20110215/c1a2f7f9/attachment-0002.html>