Sep 24, 2007
- Dan Engh and Paul Sheldon (Vanderbilt Physics)
- PR Blackwell and Diana Gunter (SFASU)
- John Cobb (Oak Ridge National Lab)
- Terry Moore (Tennessee, but calling from Opryland)
- Alan Tackett, Santi de Ledesma (Vanderbilt ACCRE)
- Micah Beck (Tennessee)
Report from PR on Arkansas Meeting
They want to see results. demo went ok, but what we need is a production level environment they can play with and use. Can't really move forward until a production level or at least pre-production level environment is available.
Deployment Plans (Alan)
summarize what was decided at our meeting Weds afternoon: what hardware and where, todo list and who is to do them, rundown of what we would like to show at SuperComputing.
Purchase Order is being prepared to purchase on the order of 50-60 depot nodes. These will put, in units of 8, at: Caltech, Florida, Michigan, SDSC, and possibly ORNL (depends of subsequent conversation) and FIU.
Already have several at Vandy, 2 at SFASU, and UCSB has their own.
Plan is to make this a "production" or pre-production facility that AmericaView (and others) users could use. Need this in place by mid-November, or at least a good portion of it.
Oak Ridge (John Cobb) needs a plan. they have space. network location would be in TeraGrid environment. mesh very will with SDSC.
Greg Pike is the coordination person. Need a date and a task list, paragraph of what we plan to do. Audience is his colleagues.
REDDnet engagement with the TeraGrid: REDDnet is going to start experimenting with folks on the Teragrid, to do that we are putting depots at SDSC and ORNL. Part of our effort will involve performance testing across the TeraGrid link, but we are also may do experiments with people on the OSG. Relevant projects currently being supported by the Teragrid are CMS, the Terascale Supernova Initiative, and FACIT, a testbed being constructed for the Library of Congress with the NGDA (National Geospatial Data Archive). Need to give some introduction about what REDDnet is, and mention that technology is currently being developed (LORS/L-Store backend for GridFTP) that will enable TEragrid projects to access REDDnet.
What kind of accessibility will these depots have to networks such as I2. Can test with a traceroute to tg-login.ornl.teragrid.org or www.ornl.teragrid.org
Just need connectivity to I2. May not be full 10 Gbs but connectivity should exist.
NSF has blessed this hosting of REDDnet nodes.
Date would be mid-October.
What we want to show: use these fairly large sites to do three things.
- Data un-tethered model of CMS analysis
- migrate data around sites - replicate data from ORNL to SDSC, or migrate... (Library of Congress)
- ability to use FIFOs... stage the data on its way...
could put some hardware at ORNL booth...
might need to schedule the network for any SC demos
send him some slides on REDDnet for his powerpoint presentation in the booth?
Equipment sent to SFASU have arrived, are being racked now, and should be online this afternoon!
Report from Washington (Terry)
Storage Vendor's meeting, gave a presentation at the end. Room populated with storage vendors. This year: security arguments. LoC people concerned that security can be counter-productive to long term. They like our minimalist approach, and don't see us as a threat to them. They have no concept of federated storage.
As near as Terry can tell, we are the theory that the LoC has for federated storage, no alternatives except vendor solutions that are not really solutions.
Report on CMS Progress (Dan)
Has the gridftp with LORS backup working (simple reads and writes 5 MB/s) stepping point for more sophisticated multi-threaded version. has HPSS template for that, needs to work with Chris Sellers. CMS PhEDEx is working at Vanderbilt, plan to eventually migrate to using this new gridftp server. Needs to talk with ACCRE guys about how to also convert that to an L-Store version.