Computing on Demand

1
Computing on Demand

• February 5, 2004

2
Status of Condor at RACF

• Condor 6.6.0 installed on all 1097 Linux Farm
  nodes in RACF.
• Standard batch policy in place (fair-share, job
  suspension, etc).
• Multiple queues for CAS CRS, sub-divided by
  experiments.

3
Changes Since Initial Install

• Graphical package (condorview) installed and
  running. Needs external Web server access.
• Strong authentication for Condor in progress
  (Kerberos and GLOBUS certificates).
• Fail-over feature to be available soon?
• Computing on Demand (COD) feature enabled.

4
Computing on Demand

• COD allows users to submit high-priority jobs
  with interactive-like response.
• COD jobs do not follow standard Condor batch
  policies (jobs run immediately after submission).
• COD jobs suspend standard Condor LSF jobs.
• COD job description file (jdf) is different from
  standard Condor jdfs.

5
Computing on Demand (cont.)

• COD only available to pre-defined list of
  privileged users.
• Currently only the reco accounts are in the
  list.
• Request for additional users with COD privilege
  must originate from RACF Liaison.

6
Computing on Demand (cont.)

• Privileged users must first request COD resources
  (see below)
• gt condor_cod request -name lthostnamegt -pool
  condor01.rcf.bnl.govltportgt -classad cod.id
• gt Successfully sent CA_REQUEST_CLAIM to startd at
  lt130.199.206.5132787gt
• gt Result ClassAd written to cod.id
• gt ID of new claim is lt130.199.206.5132787gt1071
  8567591

7
Computing on Demand (cont.)

• Hostname is FQDN (Fully Qualified Domain Name).
• Ports (optional parameter) are as follows
• 
  9660 ? ATLAS
• 
  9661 ? BRAHMS
• 
  9662 ? PHENIX
• 
  9663 ? PHOBOS
• 
  9664 ? STAR
• 
  9665 ? RCF

8
Computing on Demand (cont.)

• Next, privileged user must activate his/her COD
  claim (which runs the COD job contained in
  cod.job)
• gt condor_cod activate -id lt130.199.206.5132787gt
  10718567591 -jobad cod.job
• gt Successfully sent CA_ACTIVATE_CLAIM to startd
  at lt130.199.206.5132787gt

9
Computing on Demand (cont.)

• Users can check on progress of their COD jobs
  with
• gt condor_status -pool condor01.rcf.bnl.govltport
  gt -cod
• Name ID ClaimState TimeInState
  RemoteUser JobId Keyword
• gt vm2_at_rcas6001 COD1 Running 0000646
  starreco 1.0
• Total Idle Running Suspended
  Vacating Killing
• gt INTEL/LINUX 1 0 1 0 0 0
• gt TOTAL 1 0 1 0 0 0

10
Computing on Demand (cont.)

• Privileged users can release COD resource upon
  completion of their COD jobs
• gt condor_cod release -pool condor01.rcf.bnl.gov
  ltportgt -id lt130.199.206.5132787gt10718567591
• gt Successfully sent CA_RELEASE_CLAIM to startd
  at lt130.199.206.5132787gt
• gt Status of claim when it was released Idle

11
Computing on Demand (cont.)

• Examples of COD jdfs can be found in
• /usatlas/u/atlareco/condor/cod.job
• /brahms/u/bramreco/condor/cod.job
• /phenix/u/phnxreco/condor/cod.job
• /phobos/u/phobreco/condor/cod.job
• /star/u/starreco/condor/cod.job

12
Whats next?

• Plan is to implement COD this week on fastest CAS
  servers within each experiment.
• Maybe also implement on CRS servers for new CRS
  Batch software?
• Attend Condor Week (April 15-16, 2004) in Madison
  to learn how to improve our Condor pool. We have
  many questions