May 7, 2008

REDDnet Collaboration Meeting

• May 7, 2008 -- 11:00ET/10:00CT/8:00PT
• Call 510-665-5437
• Meeting ID is 7333

Attending

• Santi, Alan, Bobby, Matt (Vanderbilt ACCRE)
• Paul, Dan (Vanderbilt Physics)
• PR, Diana (SFASU)
• Micah, Terry, Harrold and Chris (UTK)
• Hunter (Nevoa)

Agenda

• PerfSONAR integration
• NFU proposal (Chris and Dan)
  • Important to have a vector read of some kind to reduce latency in CMS application. A couple of different options - an NFU and/or a READV call that would be part of an IPB Client.
  • have a prototype NFU which uses the UTK NFU server
  • the NFU will be detachable
  • NFU will run on separate hardware at depot sites - this is an NFU server
  • but will continue to be integrated into LoCI Depot
  • will use Hunter's NFU code to integrate - Hunter's server code
  • Chris and Dan are writing oplets and client code
  • deployment - get NFU servers up at half the sites in 1-2 months (according to Bobby)
  • could also get one maybe quicker by using one of the ORNL depots for this purpose
  • need an NFU experimental set of depots as a part of REDDnet?

• LOC Status (Santi)

• VTNA Status, meeting with Marshall and others (Santi)
  • working on automation process with creation of manifests that they want (upload and download)

• AmericaView (PR)
  • separate phone call to develop a plan of action

• TSI Progress (Terry/Chris)
  • An important question here is what the status of the ORNL depots is, since John B. will want to be able to upload to REDDnet from EWOK. If local REDDnet hardware is available, it's likely that the update will be much faster.
  • they should use depots at UTK - suggestion from Micah
  • REDDnet depots are attached to TeraGrid, outside firewall
  • Matt will start working on this now, Bobby can make it a priority starting May 19
  • in the meantime, use UTK depots

• CMS Progress (Engh)
  • GridFTP L-server using LORS and LODN seems to be working.
  • Performance? Getting about 150 Mbs/file which is the same as the GridFTP server itself (Dan thinks this is a limitation of GridFTP, not his backend). This was from Fermilab to here.
  • Should work with Bobby to see if this could be improved...
  • Plan is to have gridftp write to a local depot, then have warmer move file around.
  • Root plugin will at first use local depots for performance issues.
  • this summer put in features for vector read. could use NFU or READV and try both.
  • NFU could provide value added for CMS users

• Do we want to contribute to the OSG Operations Toolkit?