LHCb computing model and the planned exploitation of the GRID

1
LHCb computing model and the planned exploitation
of the GRID

• Eric van Herwijnen, Frank Harris
• Monday, 17 July 2000

2
Overview

• LHCb computing model and existing resources
• EU GRID project participation
• GRID unique opportunity to unify CPU resources in
  the collaborating institutes
• Mini project
• EU GRID project submitted
• Paper by F. Harris at
• http//lhcb.cern.ch/computing/hoff-grid/compmodela
  ndgrid.doc

3
LHCb computing model

• RICH and VertexDectector require detailed
  optimisation studies
• High level triggers and physics studies require
  large amounts of simulated background and channel
  data
• Simulation largely (2/3) to be done outside CERN
  (1/3)
• GRID to be used with existing FTN based
  simulation program

4
Current resources for MC production

• CERN PCSF, 60 queues running WNT
• RAL 30 queues running WNT
• Liverpool 300 PC farm running Linux
• Short term objective move away from WNT to Linux
  and use the GRID

5
LHCb WP8 application(F. Harris)

• MAP Farm(300 cpu) at Liverpool to generate 107
  events over 4 months.
• Data volumes transferred between facilities
• Liverpool to RAL 3TB (RAW,ESD,AOD,TAG)
• RAL to Lyon/CERN 0.3TB (AOD and TAG)
• Lyon to LHCb inst. 0.3TB (AOD and TAG)
• RAL to LHCb inst. 100GB (ESD for sys. studies)
• Physicists run jobs at regional centre or move
  AOD TAG data to local institute and run jobs
  there.
• Also, copy ESD for 10 of events for systematic
  studies.
• Formal Production scheduled ? start 2002 to
  mid-2002 (EU schedule)
• BUT we are pushing ahead to get experience so we
  can define project requirements
• Aim for a production run by end 2000
• On basis of experience will give input on HEP
  application needs to the Middleware groups

6
GRID starting working group

• Glenn Patrick
• Chris Brew
• Frank Harris
• Ian McArthur
• Nick Brook
• Girish Patel
• Themis Bowcock
• Eric van Herwijnen
• others to join from France, Italy, Netherlands
  etc.

7
Mini project (1)

• Install Globus 1.1.3 at CERN, RAL and Liverpool
• CERN installation completed on a single Linux
  node (pcrd25.cern.ch) running Redhat 6.1, not
  available to public yet
• RAL installed but not yet tested
• MAP being installed
• Members to get access to respective sites
• Timescale 1st week in August

8
Mini project (2)

• Run SICBMC at CERN, RAL and Liverpool
• Prepare a script using Globus commands to run
  sicbmc and copy the data back to the host where
  the job was executed from
• Other partners to test the script from their
  sites
• Timescale middle of August

9
Mini project (3)

• Verify data can be shipped back to CERN and
  written onto tape
• Use globus commands
• Some scripting required to use SHIFT and update
  bookkeeping database
• Time scale end of August
• Benchmark sustained datatransfer rates

10
Integration of the GRID between CERN, RAL and MAP
(1)

• Globus
• toolkit has a C API (easy to integrate with
  Gaudi)
• commands for remotely running scripts (or
  executables), recovering data from std output,
  saving consulting metadata through LDAP
• The gains
• Everyone uses the same executable
• Everyone uses the same scripts
• Data is handled in a uniform way
• Batch system (LSF, PBS, Condor) to be discussed

11
Integration of the GRID between CERN, RAL and MAP
(2)

• Explore the use of LDAP for bookkeeping database
• API in C would solve current Oracle -gt Gaudi
  interface problem
• Simplification of DB updating by MC production
• Everybody is heading in this direction
• Oracle have an LDAP server product, someone
  should investigate
• Java job submission tools should be modified to
  create Globus jobs. Timescale October (to be
  done in parallel with current NT production)

12
Integration of the GRID between CERN, RAL and MAP
(3)

• RAL NT farm to be converted to Linux this autumn
• MAP uses Linux
• CERN have been asked to install Globus on the
  Linux batch service (LSF)
• A full MC production run using Globus is aimed
  for in december

13
Extension to other institutes

• Establish a GRID architecture
• Intel PCs running Linux Redhat 6.1
• LSF, PBS or Condor for job scheduling
• Globus 1.1.3 for managing the GRID
• LDAP for our bookkeeping database
• java tools for connecting production, bookkeeping
  GRID