Attendees
Ray Plante
Cristina Beldica
Don Dossa
Jacek Becla
Jeff Kantor
Tim Axelrod
Robyn Allsman

Management Topics:
JK - SLAC Director's Review ­ Jeff and Tim will be at SLAC on Wednesday,
Thursday next.  Review is hurdle to getting DOE funding for camera.
JK - Cost Estimates must be provided by May, will need updated UML model as
input. 
JK - UML Model updates needed for cost estimates, keep the model current.
JK - Schema merge ­ Agreed to do, capture results in EA, focus on parts
needed for DC1 first.
JK - Working Group meetings as videocons are encouraged, to save on
telecommunications line costs.

DC1 Planning:
Two tracks within Perseus:
RP - Track 2 ­ Address throughput, scalability.  Data flow from
mountain-base-archive, with nightly processing at 2% of LSST anticipated.
Use 3 sites of TeraGrid, each represents an LSST facility.  Deploy parts of
prototype system at each site.  At base, run image calibration and object
detection pipelines.  Then move to archive center, run same pipelines, and
ingest catalog data into database.  Input data would be simulated data from
sim group.  Contingency for data ???  Will incorporate ³load generators² for
pipelines that we won¹t have.
TA ­ Started on deployment diagram for DC1.  Need to identify locations,
characteristics of the nodes on the Teragrid.  Also need to understand the
middleware that will be used for DC1.
RP ­ NCSA will put together description of infrastructure that we will use.
In Teragrid there is scratch disk space, nodes that you can reserve via
batch queueing system.  In baseline, you co-reserve nodes with Teragrid,
then you deploy, then execute, move results to permanent storage, clean up.
Also have the NCSA cluster as the Archive Node for DES CD1, but it is old.
Alternative is to get Teragrid-attached disk space.
TA - Let¹s start to exercise the Teragrid with load generators asap.  Will
expose lots of issues.
RA ­ Need to understand how to acquire resources, data.  First use would be
to test out load generators.
RP ­ You write applications as if file is on local disk, you will be passed
path to that local file.
JK ­ This is the file access API, needs to be taken to robustness diagram
level in skunk works session or sooner.  High priority for DC1.
TA ­ Put as much as you can on the table before the skunk works.
JB ­ Use cases in UML covering ingest, show how API will look.  Need to
decide on format of data.  Test single steam in March (will know
throughput/nodes needed), parallel/multiple streams in April.  May, June,
July do on Teragrid.
RP - The database engine can be deployed on Teragrid.  Several nodes may
need to run, close to pipeline nodes.
Also important to press into PCS at skunk works, at least for targeting step
selecting input/output, policies to apply for execution.  Will expand
deployment diagrams with use of Teragrid HW and document protocol
stacks/middleware into DC1 plan.  Schedule training for developers.
JK - Central cvs repository at NCSA, LSSTC needed for DC1.
RA, RP ­ Put together plan for code repository for DC1





_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data

Reply via email to