TeraGrid Arch MeetingRP Update: ORNL
January 15, 2008 (MLK + 80y)
John W. Cobb
Outline
• SNS Status• Neutron Science Portal• Data Storage Experience• Wider TeraGrid integration with Neutron Science
Portal• NSTG Cluster Operations
• Move the data: Support the experiment: connect to CI
SNS Status
• Ongoing instrument run times (SNS and HFIR)• New instruments in commissioning, joining user
program (Beamline: Name)– 2 Backscattering Spectrometer (BASIS)– 3 * Spallation Neutrons and Pressure Diffractometer (SNAP)– 4 A Magnetism Reflectometer (MR)– 4 B Liquids Reflectometer (LR)– 5 *Cold Neutron Chopper Spectrometer (CNCS)– 18 * Wide Angular-Range Chopper Spectrometer (ARCS)
• Instruments in Commissioning (shutters opened)– 6 Extended Q-Range Small-Angle Neutron Scattering Diffractometer (EQ-SANS)– 11A Powder Diffractometer (POWGEN)– 17 Fine-Resolution Fermi Chopper Spectrometer (SEQUOIA)
• More in the pipeline (10 more at last count) – “shovel ready”
Neutron Science Portal
• http://neutrons.ornl.gov/portal/ (SNS with NSTG)• Holiday success: Over break:
only 3 files failed to automaticallymove from Data Acq to portal archive.“I used the portal and it really works”
Data Storage Experience
Wider TeraGrid integration with NS Portal
• DOE SBIR with TechX corp (Mark Green) thick client portal and “virtual file system” for remote replication of SNS data (dist. Storage and disconnected use cases)
• Examining SNS data replication on archive storage targets. SULI semester Student David Speirs is also working here
• Jimmy Neutron Community Account continues use• TG dist. Jobs
–Simulation tab– Generalized fitting service (DAKOTA)– Distributed Data Reduction
NSTG Cluster Operations
• GridFTP: –We are moving to unstripped GridFTP because of “adverse
interactions” between GridFTP, PVFS, and local HW – not completely specified
–Working to implement, as planned, an enhancement path to support 10gbs transfer, especially SNS<>NSTG<>TG at large
–Will remain unstripped until then
• ESG: New local ESG node in TeraGrid enclave at ORNL – undergoing installation now
• Planning to work to mount Wide Area Lustres• General NSTG cluster usage: “canary in the coal mine”• Interesting large job count experience in Dec.
StageSub (Data Butler) Update
• See related file under this agenda on TG Wiki
• Idea: For submitted jobs, separate data movement (stage-in/out) from execution
• Dedicated, (privileged) job queue for Data transfer jobs to/from scratch
• Utility: (Users)– Not waste allocation during job execution waiting for job movement. And/or– Not risk data purge before data gets stored and/or job schedculed
• Utility: (Centers)– Increased Job throughput– Increased “effective” scratch storage– Ability to manage data storage bandwidth
• How: – Batch script directives
– Parser to submit munged scripts to compute and data queues– “Watcher” to coordinate progress between compute and data queues
• Multiple file movement tools including standard center tools as well as archive and Wide area tools including P2P tools (freeloader)
• Current Deployment (and future deployment)