Re: [GlueX] computing requirements for PAC30 documentation

[Elton Smith <elton@jlab.org>]

Hi Richard,

We are going ahead and submitting the computing cover sheet with your
numbers. However, unless something has changed in the general architecture
design, we have not thought that we would be storing raw computer data on
tape at a central site. Storage could be made available if it is
necessary, but it is not cheap. These would be processed and only the
reconstructed data would be moved from one site to another.

Cheers, Elton.




Elton Smith
Jefferson Lab
elton@jlab.org
(757) 269-7625

On Thu, 6 Jul 2006, Richard Jones wrote:

> Alex, David,
>
> Following a lengthy discussion during the software meeting, I agreed to consolidate my notes into a table
> and distribute them.  I believe that this represents the state of the knowledge of our requirements.
> Note that these numbers are VERY different from those shown on the first draft that David sent around.
> My results are summarized in the attached spreadsheet.  In particular, note the following points.
>  1. David's estimate for MC cpu time was based on his SPECint2K ratings.  The document asks for SPECint95
>     units.  The two are different by more than an order of magnitude.  See the attached spreadsheet for
>     details.
>  2. David's estimate for MC event count (3x10^11 events) taken from the design report is computed based
>     on statistics for the entire experiment, including 10^8 tagged g/s running.  For the first two years
>     at 10^7 g/s the total is smaller by about an order of magnitude.
>  3. The "On-Line Disk Storage Required" line is ambiguous.  Does "online" mean data acquisition staging
>     area (a FIFO for the silo writer) or does it mean the total staging area needed for the calibration,
>     reconstruction, and analysis activities on-site? The former is a very small subset of the latter.  I
>     assume the latter. If I am wrong then where on the form are we supposed to specify our requirements
>     for work disk?
>  4. For network import capacity, I assume that it is dominated by the need to archive the generated MC
>     datasets in a central location.  Everything else is in the noise.
>  5. For network export capacity, I assume that it is dominated by outside institutions grabbing DSTs for
>     final cuts selection and PWA.  That is how I computed it.  I do not see the sense in having outside
>     institutions import the entire raw data sample.
> Richard Jones
>
>