Use And Implementation of EGEE Grid by CMS

1
Use And Implementationof EGEE Grid by CMS
LCG POB Meeting June 20, 2005
2
CMS MC Production on EGEE Grid

• All CMS MC production is Grid based
• on EGEE Grid concentrating on Tier-1 sites
• running full MC production chain
• struggling to achieve goal of 10M events / month
• significant trouble to get smooth running
• main roadblock site mis-configurations, lack of
  stability and robustness, slow human interfaces
• collecting quantitative evidence of successes and
  failures
• Experience shows that CMS requires a fair number
  of dedicated staff for production running on EGEE
  Grid (2-3 FTE/Tier-1 as requested in MO-B)

3
CMS User Analysis on EGEE Grid

• Data analysis for CMS Physics TDR, very visible
  activity in CMS!
• sites host CMS data sets serving data to user
  analysis jobs
• requires production-quality stability,
  throughput, user support
• CMS data on EGEE Grid accessible to users through
  CRAB Grid interface
• CRAB system for running CMS analysis jobs at
  sites that host CMS datasets
• at CERN, Fermilab, smaller samples at Tier-2
  sites (mainly INFN Tier-2s).
• CMS will work with more Tier-1 centers for
  hosting of more dataset
• not all Tier-1s seem ready for hosting large CMS
  datasets serving many users
• also, very high throughput required for analysis
  running (300MB/sec for 20 CMS users, i-e same
  as CDF or D0)
• CMS Computing to prepare an initial Operations
  Model and Plan
• clarify responsibilities of user - CMS-VO - LCG -
  EGEE/OSG - site
• define/explain approaches to incident handling
  and communications paths
• clarify user support, expectations on quality of
  service etc

4
Integration and Service Challenges

• EGEE/gLite components listed as candidates for
  Baseline Services
• no experience exists in CMS with most of the new
  components
• large integration effort required for new
  components to work for CMS
• require interoperability in approaches and
  interfaces between all grids
• SC planning must address the introduction of
  these new components
• a clear roadmap needs to be developed and agreed
  on
• also address possible incompatibilities with
  existing services
• CMS committed to get fully involved in SCs on
  EGEE and US grids
• CMS has created a dedicated integration program
• require EGEE/LCG integration program, involving
  EGEE developers
• require pre-production environment for to-scale
  end-to-end testing
• (e.g. run on 1000 CPUs in test mode, as part of
  SC exercise)

5
Backup Slides on Status and Assessment of CMS
Work in ARDA
6
gLite Components Becoming Available
7
ARDA Prototype CMS Analysis

• Application demonstrator ASAP
• end-to-end prototype for distributed analysis
• Running on ARDA testbed at CERN, real data
  analysis running, selected users
• Puts some of the gLite components to work on the
  gLite testbed at CERN and UWMadison
• Uses and exercises WMS (gLite workload management
  system)
• Monitoring and job control services, including
  gLite and MonALISA
• CMS data management components (refDB, pubDB)
• Some good experience gained using gLite
  middleware
• 90-95 success rate for hello world type
  applications, but ups and downs
• also done performance test on gLite file
  catalog/manager
• Fireman catalog, comparison to LCG file catalog
• other components like FTS not used because
  testbed not distributed
• Showed some current limitations of the gLite
  prototype and giving feedback to gLite developers

8
Some successes with gLite and ARDA

• This is a very useful and successful start in
  using gLite for CMS analysis
• Also other CMS components (BOSS) interfaced to
  gLite already
• However, we should be realistic about the scope
  of this exercise
• very limited and unrealistic setup
• not distributed, few nodes, no access to large
  s of WNs
• not scoped to expose the main current issues with
  LCG-2 related to the highly distributed nature of
  the environment and responsibilities
• configuration stability, scalability, robustness,
  clear definition of responsibilities and
  expectations etc
• CMS needs to move this work into the main line of
  CMS integration and reconcile with CMS Computing
  baseline