Application Parallel Session for OSG 0.4

1
Application Parallel Sessionfor OSG 0.4
Frank Würthwein UCSD
2
Purpose of this discussion

• Bring sites applications together to talk about
  whats in OSG 0.4
• Focus on 2 themes
• Storage
• Reliability

3
Storage

• Kent on LIGOs needs
• Ian on SRM/dCache
• Eric on SRM/DRM STAR
• Gaurang on VDS/Pegasus

4
Reliability

• What sort of experience do apps have?
• What sort of experience do sites have?
• What are we going to do about it?

5
Problems site perspective

• The way pre-ws models its state in forks
• Apps need to always use grid monitor!
• Streaming of output
• Apps get used to turning this off
• Will allow sites to reject jobs that turn on
  streaming!
• Gftp server on the CE
• Sites should separate CE SE on different
  hardware
• Jobmanager-fork
• Will have schedd and local universe behind
  jobmanager-fork

6
Job output handling

• a cluster is a large error amplifier
• Make sure you clean up after yourself!
• Copy out what you want back using srmcp rather
  than depending on globus/condor stage-out because
  srmcp does traffic shaping, and thus does not
  overload the CE!
• See e.g. STAR workflow described by Eric!