Use of Condor on the Open Science Grid

1
Use of Condor on the Open Science Grid

• Chris Green, OSG User Group / FNAL
• Condor Week, April 30 2008

2
What is OSG?

• Links
• OSG home page.
• VORS resource map and information.
• VDT (Virtual Data Toolkit) home page.
• Current use of OSG.

• Collection of mostly US-based scientific /
  academic sites sharing computing and storage
  resources via common software stack.
• Job submission and management based around Globus
  / CondorG.

• "Virtual Organizations" (VOs) trust point for
  authorization role-based personalities.
• Works with multiple underlying batch systems
  (Condor, PBS family, LSF, SGE).

3
OSG facts and figures

• 83 registered computing resources.
• 30 registered VOs.
• Usage breakdown for 2008/04/19 2008/04/25

4
Survey of Condor useon OSG

• Out of the box
• CondorG for inter-site job transfer via
  Globus/GRAM GT2 submissions via CondorG still
  (by far) the most common method of grid job
  submission on OSG.
• Task scheduling for site health monitoring.
• One of several batch systems supported on OSG.
• "ManagedFork" job management.

5
Survey of Condor useon OSG

• External projects
• Glidein / WMS "pilot" job submission and
  management.
• FermiGrid job forwarding, "campus grid"
  management.
• OSGMM / ReSS job forwarding and attribute-based
  matchmaking across multiple OSG sites.
• "condorview" enhanced job monitoring and control
  not the web-based statistics client of the same
  name.
• Complex workflows (eg LIGO Pegasus/DAGMAN).
• Gratia accounting system leverages features of
  condor where available condor_history,
  PER_JOB_HISTORY_DIR, DN.

6
More detail Glidein/WMS

• Workload Management System (Igor Sfiligoi, FNAL)
  uses Condor Glideins -- startd submitted as a
  grid job ("pilot") makes remote batch nodes look
  like local ones.
• Two main components
• One or more glidein factories manage available
  grid sites and submit pilot jobs.
• One or more VO frontends receive payload
  submissions from users for distribution to sites.
• Pilots receive user payloads as distributed by VO
  frontends.

7
More detail Glidein/WMS
8
More detail Glidein/WMS

• Uses GCB for firewall / NAT management .
• Intra-VO priority management.
• Works with glExec application running on worker
  nodes which handles authorization and UID mapping
  for payloads per user accountability to the
  site.
• Unaffected by grid site batch manager choice.
• V1.0 released Dec.'07 v1.1 Jan'08.
• In use by CDF Minos (FNAL) being commissioned
  for CMS.

9
More detail "condorview"

• Michael Thomas, Caltech.
• Graphical tool for browsing and managing a condor
  queue.
• Hooks to vacate and kill jobs.
• Hooks to ssh into job directory on worker node
  and print out process tree.
• Uses condor_q, condor_config_val, and
  condor_fetchlog.

10
More detail condorview
11
More detail condorview
12
Concluding statements

• Condor essential to the OSG.
• Condor use underpins connectivity of sites within
  the OSG.
• Close ties Miron is OSG PI VDT team at
  Wisconsin new Condor features often a result of
  OSG needs.
• Widely used on OSG many novel uses of and
  applications building on Condor features.
• More details in later talks!