[Halld-offline] Offline computing plan for JLab Physics, FY10

[Richard Jones]

Chip,

I fully agree with this requirements-driven approach.  It is the way we 
achieve and justify designs of experimental apparatus.

This message from me that you responded to was following up on a 
discussion that took place during the collaboration meeting earlier this 
week, where I brought up the concern that offsite access to data at Jlab 
is (in my experience as a user) antequated and laborious.  My concern 
was that our own (GlueX) offline computing plan has not clearly 
articulated our requirements for ease of offsite access to data at Jlab, 
and that if we do not write it in black and white in our requirements, 
it does no good for us to ASSUME that it is just going to happen 
automatically.  If it is not written in the requirements, we have no 
grounds to expect it to be there.

Over the past decade we have read in CC documentation that the SRM tool 
would be coming soon to meet this need.  However we have learned 
recently that support at the laboratory for offsite SRM access to MSS 
has atrophied.  No doubt this is the result of strain on existing 
resources and lack of a clear message from users that this capability is 
needed.  I asked Sandy why support for SRM is going away, with GlueX 
depending on it, and she pointed out that there is no mention of it in 
our offline computing plan. Touche.

With these cost estimates in mind as ballpark figures, we are better 
able to have this conversation as a collaboration and flesh out what our 
requirements are for offsite access.

-Richard J.

> Richard,
>
> The cost of the items you are recommending is significant:
>
> 1.  The cost of developing location-independent access to MSS files is 
> probably an additional $100K above current capabilities, perhaps as 
> much as twice that depending upon details of requirements.
>
> 2.  The cost of storage is independent of where the work is done -- 
> the full requirements are meant to be in the 12 GeV Computing Plan, 
> and if you don't believe the requirements are correctly captured, 
> please work with Graham Heyes to correct that.
>
> 3.  The cost of maintaining such a full computational grid capability 
> is approximately $180K per year for staff, plus some modest hardware 
> costs for appropriately provisioned gateways, depending upon detailed 
> requirements.  (And if tier-N sites were to appear, they would also 
> require grid administrative overhead beyond JLab's costs.)
>
> Jefferson Lab is committed to satisfying the requirements for 12 GeV, 
> but does not start a priori with an assumption of a multi-tiered grid 
> -- that is an implementation choice, not a requirement.  There are a 
> number of political and scientific reasons why LHC made that choice, 
> many of which are not valid for Jefferson Lab.  The current 12 GeV 
> requirements do not drive us toward a computational grid solution, and 
> so the laboratory has as its working plan a much more cost effective 
> mostly centralized solution.  Of course if the requirements are not 
> correct, the solution may not be correct, but any significant change 
> in the scale of the requirements will have to compete for funds with 
> detector enhancements, accelerator operations, and many other costs.  
> Offline computing is not part of the construction project, but only 
> the operations budget.
>
> I began the process of requirements collection so that Jefferson Lab 
> could move in the direction of a requirements driven solution, with 
> sufficient lead time so that it would not be left until the end and 
> thus under provisioned.  HEP has a history of long lead time 
> activities, with LHC being a premier example.  Bear in mind, however, 
> that our requirements are much less than 1% of LHC's, and our budget 
> is correspondingly much smaller.
>
> regards,
> Chip
>
>
> Richard Jones wrote:
>> Offline folks,
>>
>> One thing stands out to me in reading through this plan is the lack 
>> of any accommodation for the  interaction between the Jlab computer 
>> center and computing facilities at user institutions.  For 
>> comparison, it is helpful to look at the computing plan for any of 
>> the LHC experiments, where the plan is oriented around distribution 
>> and global access to data and computing resources.  Things that I 
>> would like to see in the Gluex Offline Computing Plan that eventually 
>> get propagated through appropriate channels to someone like Graham 
>> Hayes are the following:
>>
>>    1. Location-independent mechanisms for remote access to MSS files
>>       (with appropriate authentication/authorization) without logging
>>       in on a Jlab machine over ssh, both reading and writing.
>>    2. Provision for allocation of space on MSS for archival of
>>       intermediate data sets generated off-site, eg. MonteCarlo and
>>       final data samples used in a PWA that was the basis of a
>>       particular publication.
>>    3. Support of grid protocols for transparent scheduling and
>>       migration of jobs across available resources within the
>>       collaboration, so that when a particular resource is overbooked
>>       jobs can flow across site boundaries to where resources are
>>       available.
>>
>> -Richard Jones
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20100515/6035dbdc/attachment-0002.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 3639 bytes
Desc: S/MIME Cryptographic Signature
URL: <https://mailman.jlab.org/pipermail/halld-offline/attachments/20100515/6035dbdc/attachment.p7s>