SA1 - Status

1
SA1 - Status

• J.Gomes, G.Borges, M.David
• LIP

Int.eu.grid integration meeting Lisbon, November
2007
2
Status

• The most needed middleware functionalities are
  now either available or under validation
• Open MPI
• PACX MPI
• MPI START
• CrossBroker
• support for all of the above
• glide-in at least partially
• Interactivity with application steering Glogin
  and Gvid
• RAS/MD

3
What do we need technically

• Focus on what we really need
• Finish the validations
• Perform thoroughly tests (new critical
  functionalities)
• ResourceBroker Glide-ins
• PACX-MPI and OpenMPI
• Interactivity
• RAS/MD
• Deploy the validated middleware in production
• Focus on fixing any pending issues
• Both at the level of the middleware and
  infrastructure
• Understand why some job types do not work
  properly
• Provide stability for the coming months
• Very important for the end-users
• No major issues
• No major changes in production
• Meet the sites resources committed in the
  technical annex
• Support for low latency interconnects

4
What do we need to show

• Need to show that
• Users can carry true production work in the
  infrastructure
• Need to show a reasonable number of customers
• using the infrastructure systematically
• not just testing and porting
• Need to show good resource usage
• Interoperability
• We can cope with production loads
• Heavy usage may show additional issues
• Brokering and resource availability
• Sites configuration
• Other middleware bugs
• Not enough resources for the load
• We are speaking about interactivity hence quick
  response

5
Interoperability

• Considerable effort to study scenarios and
  solutions
• Focused in supporting existing VOs
• Presented at the EGEE conference in Budapest
• So far without tangible results
• One result we have shown that it can be done
• Are we interoperable with gLite ?
• The LCG RB is being abandoned in favour of the
  gLite-WMS
• The gLite-CE was abandoned in favour of CREAM
• The LCG CE should around for some time (ported to
  SL4)
• Classic SEs are no longer supported
• SLC3 no longer supported
• SL3 is still supported
• EGEE is moving to SL4

6
Monitoring and Accounting

• A lot of work done to provide
• Monitoring tools
• GridICE, site admin notifications
• Site tests
• SAM now with tests for OpenMPI, PACX-MPI and soon
  interactivity
• Check installed packages
• Accounting portal
• With support for MPI and VO specific tests
• Useful tools
• Need to check the accuracy of the test results
• Need the help from the site admins and everybody
  for this
• Fundamental, a lot of these verifications are
  being performed manually

7
Middleware support and deployment

• Development cycle
• Development-gtValidation-gtProduction deployment
• Repository and autobuild at FZK stable
• Need autobuild support for SL4
• Drop the support for Itanium
• The new cluster at PSNC will be x86
• It will simplify significantly the overall
  infrastructure and services
• Forget about non x86 architectures ?

8

• Fix PACX-MPI issues at most sites
• Fix interactivity issues
• Get the new CrossBroker in production
• Test the glide-in approach in production
• Get new RAS server in production
• Drop the classic SEs and move them to DPM
• the sites with RAS servers must also operate a
  classic SE.
• Deploy an immediate job startup mechanism for PBS
• Consider to move most sites to the pbs-jobmanager
  
• lcgpbs much slower

9
VO usage
10
VO usage