MinutesOct23

Introduction

• Attendees: Steve Timm, Arvind, Charles Bacon, Tim, Anand, Jeff Porter, RobQ?, Stu, Horst, Karthik
• Apologies: none
• Coordinates: Tuesdays, 3:30pm Central; 510-665-5437, #1212
• Previous meetings, MeetingMinutes
• ITB 0.7 background
  • ReleaseDocumentation
  • SiteValidationTable

Changes to attributes / schema

• Steven is seeing an interesting problem with fermigrid:
  • Fermigrid has multiple heterogeneous subclusters
  • Number of job slots gets multiplied by subclusters resulting in overreporting of # of free cpus available
  • Matching on an individual node in a specific subcluster is difficult to do (matching on CEs and subclusters might help here)
• Other issues (site specific services, e.g. ws-gram, nfs lite):
  • Possibly need extensions to the glue schema
  • GlueService? attributes might work
  • Need to investigate whether these will work or whether it'll break the bdii (Steven will investigate this)

Future plans for attributes subcommittee

• Should reconvene about a month or so after the next release to discuss changes/new attributes
• Somewhat of a requirements gathering process

BDII server issues (RobQ?)

• Four options to deal with issues, will be discussed in today's facility meeting
• No decision but there are several options
• Running multiple services will fragment information onto different servers
• Rob will see what EGEE is doing in with schema updates

WSGRAM testing progress (Jeff, Suchandra)

• Jeff's report
  • ran thru Validation steps against persistent ITB set
    • 2 sites had config issues - both now fixed
    • 2 sites with permission issues - unresolved
    • remainder were successful
  • began scaling tests from ITB (vdt-181) submit host w/ gsiftp installed from vdt:Globus-Base-Data-Server
    • simple Condor-g submissions against FNAL, TTU, BNL, & LBNL services
    • scaling up to 100 jobs per submit-file (condor -version 6.8.6, configured with defaults)
  • results were generally successful
    • small fraction of jobs fail with "Staging error for RSL element" (FNAL & BNL)
    • Large load on LBNL service (reaching 20) causing a larger number of "Staging errors for RSL element"
    • Need to apply performance recommendations and ask rest of ITB group
• Suchandra will start testing this week

VO WS-GRAM validation (Jeff, Suchandra)

• Jeff's report
  • Working with STAR admin (Wayne) and Grid developer (Leve)
  • Using SUMS (STAR job scheduler) built appropriate job files and condor-g submit file for gt2
  • modified submit file for gt4
  • modified job files for submission from my ITB client (STAR production clients currently do not support gt4)
  • ran jobs against FNAL, TTU, BNL - so far each have had distinct errors
    • shared lib missing on FNAL - now fixed and rerunning
    • TTU - source $OSG_GRID/setup.csh produced an error
    • BNL - gridftp of results back to submit host failed
• Suchandra's report
  • Britta (LIGO) has a ws-gram enabled workflow
  • Has submitted against UC_ITB but ran into problems
  • Resubmitting to UC_ITB and submitting to other permanent ITB sites (Fermigrid and BNL)
  • Need LIGO
• VOs have concerns in regards to the version of the osg-client package, using ws-gram depends on a newer version of the client than distributed with 0.6.0
• Uptake might have to wait until osg 0.8.0 due to problems with condor 6.8.2/6.8.3 and firewalls

Last minute changes

• Repositories will be frozen today so any changes should be given to Suchandra ASAP

AOB

• None