London Tier 2

1
London Tier 2

• Status Report
• GridPP 13, Durham, 4th July 2005
• Owen Maroney, David Colling

2
Brunel

• 2 WN PBS _at_ LCG-2_4_0
• R-GMA and APEL installed
• RH7.3 LCFG installed
• Additional farm being installed
• SL3
• Private networked WN
• 16 nodes
• Expected to move into production after 2_6_0
  upgrade
• Hoping to bring further resources over the summer
  
• Recruiting support post with RHUL (Job offer
  made)

3
Imperial College London

• Appointment of Mona Aggarwal to GridPP Hardware
  Support Post
• 52 CPU Torque HEP farm _at_ LCG-2_5_0
• RGMA and APEL installed
• OS RHEL 3
• IC HEP participating in SC3 as the UK CMS site
• dCache SRM installed with 2.6TB storage 6TB on
  order
• Another 6TB on order
• Numerous power outages (scheduled and
  unscheduled) have caused availability problems
• London e-Science Centre
• SAMGrid installed across HEP and LeSC
• Certified for D0 data reprocessing
• 186 Job Slots
• SGE farm, 64bit RHEL
• Globus-jobmanager installed
• Beta version of SGE plug-in to generic
  information provider
• Firewall issues had blocked progress but this has
  now been resolved. Testing will start soon.
• Community of Interest mailing list established
  for sites interested in SGE integration with LCG

4
Queen Mary

• 320 CPU Torque farm
• After difficulties with Fedora 2, have moved LCG
  WN to SL3
• Departure of key staff member just as LCG-2_4_0
  released led to manpower problems
• GridPP Hardware Support post filled
• Guiseppe Mazza start(ed) 1st July
• RGMA and APEL installed early in June.

5
Royal Holloway

• Little change 148 CPU Torque farm
• LCG 2_4_0
• OS SL3
• RGMA installed
• Problems with APEL default installation
• Gatekeeper and batch server on separate nodes
• Little manpower available
• Shared GridPP Hardware Support post with Brunel
  still in recruitment process
• Job offer made?

6
University College London

• UCL-HEP 20 CPU PBS farm _at_ LCG-2_4_0
• OS SL3
• RGMA installed
• Problems with APEL default installation
• Separate batch server to gatekeeper
• UCL-CCC 88 CPU Torque farm _at_ LCG-2_4_0
• OS SL3
• RGMA and APEL installed
• Main cluster is SGE farm
• interest in putting SGE farm into LCG and
  integrating nodes into single farm

7
Current site status summary
Site Service nodes Worker nodes Local network connectivity Site connectivity SRM Days SFT failed Days in scheduled maintenance
Brunel RH7.3 LCG2.4.0 RH7.3 LCG2.4.0 1Gb 100Mb No 21 16
Imperial RHEL3 LCG2.5.0 RHEL3 LCG2.5.0 1Gb 1Gb dCache 26 28
QMUL SL3 LCG2.4.0 SL3 LCG2.4.0 1Gb 100Mb No 45 12
RHUL RHEL3 LCG2.4.0 RHEL3 LCG2.4.0 1Gb 1Gb No 22 29
UCL (HEP) SL3 LCG2.4.0 SL3 2.4.0 1Gb 1Gb No 9 30
UCL (CCC) SL3 LCG2.4.0 SL3 LCG2.4.0 1Gb 1Gb No 12 9

1. Local network connectivity is that to the site SE
2. It is understood that SFT failures do not always
   result from site problems, but it is the best
   measure currently available.

8
LCG resources
Site Estimated for LCG Estimated for LCG Estimated for LCG Currently delivering to LCG Currently delivering to LCG Currently delivering to LCG
Site Total job slots CPU (kSI2K) Storage (TB) Total jobs slots CPU (kSI2K) Storage (TB)
Brunel 60 60 1 4 4 0.4
IC 66 33 16 52 26 3
QMUL 572 247 13.5 464 200 0.1
RHUL 142 167 3.2 148 167 7.7
UCL 204 108 0.8 186 98 0.8
Total 1044 615 34.5 854 495 12
1) The estimated figures are those that were
projected for LCG planning purposes http//lcg-co
mputing-fabric.web.cern.ch/LCG-Computing-Fabric/GD
B_resource_infos/Summary_Institutes_2004_2005_v11.
htm 2) Current total job slots are those
reported by EGEE/LCG gstat page.
9
Resources used per VO over quarter (kSI2K hours)
Site CPU ALICE ATLAS BABAR CMS LHCB ZEUS Total
Brunel 6 149 155
Imperial 19 848 221 4,863 312 6,263
QMUL 41 116 82,697 82,854
RHUL 1,124 1,840 79 42,218 45,261
UCL 6,982 126 14,115 21,223
Total 1,143 9,711 548 144,042 312 155,756
Data taken from APEL
10
Njobs percentage numbers of jobs
Expressed as a pie chart
51,209 according to APEL
11
Site Experiences

• LCG-2_4_0 release was first scheduled release
  date
• Despite a slippage of 1 week in the release (and
  an overlap with EGEE conference) all LT2 sites
  upgraded within 3 weeks
• Some configuration problems for a week after
• Overall experience was better than the past
• Farms are not fully utilised
• This is true of the grid as a whole
• Will extend the range of VOs supported
• Overall improvement in Scheduled Downtime (SD)
  compared to previous quarter.
• QMUL had manpower problems
• NB Although QMUL had highest number of (SFT
  failureSD) provided most actual processing power
  during quarter!
• IC had several scheduled power outages, plus two
  unscheduled power failures
• Caused knock-on failures for sites using BDII
  hosted at IC
• IC installed dCache SRM in preparation for SC3
• Installation configuration not simple default
  configuration was not suitable for most Tier 2
  sites and changing from the default was hard
• Some security concerns installations not Secure
  by Default
• Coordinator- Owen Leaving in two weeks
• Have made an offer