• CEPH Validation checklist
  • Works with all ROOT operations - opening, merging and updating file -- doesn't work in HDFS, works in CEPH
  • Running CMSSW jobs -- test stage-out
  • GridFTP transfers
  • Checksum calculation for standard Linux tools
  • Test Caching policy and functionality -- Cache node disks have to serve as a RAID0.
  • Partial replication -- less than factor of 2 usable space.

Necessary - ASAP

• Memory review/purchase -- How much memory the HT capable nodes need. Aim for 3 GB per core. 2 would be ok too.
• 2014 Deployment plan -- power / placement
  • Get quotes for Nodes/Disks. Compare 6378 with 6382 SE
  • Before finalizing it would be nice to have
• Monitoring review - test SMS alerts and fine-tune priorities of alerts and where they go
• PhEDEx migration to better hardware - will support thousands of parallel transfers instead of hundreds.

Necessary - Long term

• Write CMSSW benchmark suite that could be ran by Condor/CRAB jobs. Compare performance of whole-node to 1 core.
  • Bonus feature - Have CouchDB centralizing jobs outputs. We would know how much EV/s each CPU could do. Automated systems such as WMAgent could use this information.
• Not so long term -- Figure out the HEP Cluster usage.

Enhancement

• Start a Cloud site with the $500 Google Cloud promotion. Needs quite some testing. Setting up a minimalistic site would be fast though. Might have problems in the Middleware.
• CRAB submitting to the T2 through HTCondor CE.
• T3 Storage - CEPH commissioning, start testing with WN storage.
• Try registering T3-Higgs as T3_US_Caltech and mimic the T2 activities. Could observe how a "CEPH site" would behave.
  • In principle no mistery, if file reading by jobs and GridFTP works. Already tested ROOT access. Works better than Hadoop (support merging natively)

Computing Topics

Intended to contain all topics to be discussed in meetings. Between an agenda and a notebook.

/raidX deprecation and new homes

Pretty complex plan described in the thread "Migration of $HOME/raidX to DFS". The idea is to get rid of all RAIDs and turn them into hadoop datanodes. the $HOME area will be in an actual RAID5 + HSP. 20 TB of space.

Condor Migration

Supposed to happen in July 16th. 260 cores already in Condor. 80 on their way. SGE will be decommissioned by then.

AAA Usage - tests and instructions

• We have ComputingAAATests which is intended to be a knowledge base of the group of that works or not in terms of Xrootd access.
• In principle, works at any site that has a proper site-local-config.xml.
  • The basic procedure with CRAB is - tell it to ignore data location, whenever it runs, it won't find the data, fallback to Xrootd, query the redirector for the file and in principle run fine.
• Tested in the T3. Works as long as we have $X509_USER_PROXY defined to a location in a shared filesystem.

CRAB 3 testing

We might get some interesting suggestions of exercise for CSA14, including usage of MINIAOD. Despite of that we should evaluate/commission usage of CRAB3 by itself.

Here is the page that contain instructions for CRAB3.

-- Main.samir - 2014-07-10