JOINT EXPERIMENTATION ON SCALABLE PARALLEL PROCESSORS JESPP

1
JOINT EXPERIMENTATION ON SCALABLE PARALLEL
PROCESSORS JESPP Program Managers Review
Computational Sciences
17 Jul 07Dan M. DavisRobert F.
Lucas ddavis,rflucas_at_isi.edu
2
Overview

Contract status Last Fall, Winter and
Spring Current Efforts Outreach Backup Material
3
Overview of Contract

• Not direct sub to JFCOM, admin through AFRL
• Currently overspent
• Received April Increment
• 200K
• Brought Contract accounts to Zero (mas o menos)
• Reduced staff loadings
• Bob Lucas no charges since Spring 06
• Dan Davis On HPCMP PET Fall 06, UNR Spring
  DTO now
• Tom Gottschalk CalTech stopped work last
  summer, still comes
• Ke-Thia Yao Working projects in home Division
  DTO
• Craig Ward Programming for RFQ Editor
  SIMC-IC
• John Tran Programming for an Educ. Project
• Gene Wagenbreth Projects (UNR) till Jul, now
  DTO SIMC-IC

4
Financial and Funding Issues
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul
5
CY 07 Funding and Expenditures
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
6
Activity Last Fall-Spring

• Supported Urban Resolve phase 2
• On site at J9, SPAWAR, or TEC
• Interface to centers for Koa and Glenn
• Trouble shooting SPP and network problems
• Responsible for logging data from experiments
• Maintained jlogger S/W
• Staged data transfers to J9 after events
• Scalable Data Grid (SDG) Development
• Demonstrated early version late last summer
• Worked on new cluster design and specs

7
Leadership Outreach

• Attended several meeting in DC to discuss JFCOM
  simulation technology
• Drafted and presented paper and presentation at
  HPCMP Users group conference in PittsburghA
  GPU-Enhanced Cluster for Accelerated FMS  
• Will attend Cluster Symposium
• Drafted three papers for I/ITSEC, all selected
  for consideration of acceptance
• Will attend SC07 in Reno with Univ of Southern
  Cal.

8
Current Activities

• Weekly Phone con attendance
• Software trouble shooting and maintenance
• HPCMP documentation
• Supporting DHPI award
• Moderating Koa and Glenn termination issues
• Weekly phone/emails
• New cluster Acceptance Test Design
• GPU research

9
Software Development

• Development of SDG
• On hold since I/ITSEC
• Several outstanding issues
• Need productization for release
• Will not be ready for final release in time to
  make it into JSAF S/W release
• Converting jlogger to 64 bits (ongoing)
• Several issues resolved already
• Attacking each new issue as it is identified

10
HPCMP Documentation

• Prepared paper for Users Group Conference
• June, 2006 in Pittsburgh
• Images of UR for annual report
• Will attend Cluster Symposium
• Frequently called upon by HPCMP to find evidence
  and graphic representations of J9 success with
  HPCMP clusters

11
Supporting New DHPI

• Liaison to HPCMP
• Organized phone calls with LNXI
• Renegotiated GPU spec. with HPCMP
• Pursued price break with Nvidia (unsuccessful)
• Proposed a test plan for acceptance
• Monitoring acceptance test plan execution

12
Koa and Glenn

• Usage of SPPs is low since Oct.
• Partly due to access control changes
• Koa and Glenn now turned over to centers
• MHPCC has classified user in the wings
• Koa will disappear
• Glenns transition less definitive
• Drafted letter for JFCOM, no response

13
Seeking New Users

• Seeking other users for J9 technology
• (not spending J9 money doing so)
• NSA
• Working with Jim Heath
• Extend Urban Resolve to broader IC
• training, awareness, decision aids
• Port of Los Angeles
• Working with USC DHS center
• Proposing model of Port of LA and environs
• Educational uses

14
Organizational Diagram
15
Computing Infrastructure
Deployed, spring 04 MHPCC ASC-MSRC 2 Linux
Clusters 24x7support by HPCMP DREN
Connectivity Users in VA and CA Application
tolerates network latency Real-time interactive
supercomputing
16
Back up Slides
17
JFCOM Mesh Diagram (Notional)
18
Future Work Needed and Planned

• JFCOM has IMMEDIATE need for more entities (10X)
• Memory on Nodes and in Tertiary Storage very
  limited
• TeraByte a week
• Keeping only 20 of current data
• Need 10X more entities
• Need 10X behavior improvement
• Now doing face validity
• Need more quantitative, statistical approach
• Caltech Dr. Thomas Gottschalk
• NPS Profs Sanchez and Lucas
• Data mining efforts now commencing

19
Two Key Challenges

• Collect the fire hoses of data generated by
  large-scale distributed sensor rich environments
• Without interfering with communication
• Without interfering with simulator performance
• Maximally exploit the collected data efficiently
• Without overwhelming users
• Without losing critical content
• Goal
• Unified distributed logging/analysis
• infrastructure, helps users and the
• computing/networking infrastructure managers