Keywords: DataAccWG

Notes from DataAccWG telecon 21 Apr 2006

Attendees: Jeff Kantor, Ray Plante (moderator), Ani Thakkar

staging data use case
=====================

As part of the email discussion that followed up the introduction of
this topic on the Database telecon, Jeff pointed out that we are
planning for a 5-Gbs, full-duplex connection between the base and the
archive center.  On this note, Ray posed the general question,  do we
need to plan for efficient mechanisms for returning (database) data
from the archive center to base facilities (i.e. that minimizes
copying data that is either already there or is not needed for the
next night of observing), or can we expect to get away with simpler
approaches (i.e. just transfer everything we think we need
regardless).  

Upper limit arguments suggest that efficient strategies may not be
necessary.  If it is a requirement that we transfer all of a night's
data (15 TB) in 24 hours, then we should be able to transfer back the
new database data from that night, which is less than that (6 TB?).
The database data we actually need is likely much less than that.  So,
not considering the image data requiring caching, we should have more
than enough bandwidth.  

Separate from this are constraints on what gets cached based on the
storage capacity at the base facility.  

Discussion is limited by our understanding of database sizes: (size of
records, number of records per patch of sky, etc.).  We have some
notion of these numbers today, but it is time to collect them
together.  We need to connect with people working on scheduling issues
to see what the requirements are on data staging, and if limits on
capacity impose constraints on the scheduling (action: Ray).

event database
==============

A discussion about an event framework was started last week at the
Middleware WG; see docushare Document-1615 for a conceptual design
document.  This document touches on the event database as a
site-local, rolling cache of time-ordered events going back some
limited time into the past, primarily for the purposes of debugging. 

Jeff recommended that a useful contribution of this working group
would be to help define the kinds of events and data content
required.  In particular, we could walk through the UML model to
identify event producers and consumers.  


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

Reply via email to