JOINT EXPERIMENTATION ON SCALABLE PARALLEL PROCESSORS JESPP

1
JOINT EXPERIMENTATION ON SCALABLE PARALLEL
PROCESSORS JESPP Program Management Review
Computational Sciences
16 Aug 07Dan M. DavisRobert F.
Lucas ddavis,rflucas_at_isi.edu
2

• JFCOM J-9 MS
• Program Management
• Review
• 16 August, 2007

3
Agenda

• Personnel
• Accomplishments / Projected Tasking
• Fiscal Status
• Wrap up
• Issues

4
JFCOM Support Personnel

• Principal Investigator Robert F. Lucas, PhD
• Program Manager Dan M. Davis
• Technical Team Thomas D. Gottschalk, PhD
  (Caltech), Ke-Thia Yao, PhD, Gene Wagenbreth,
  John J. Tran, Craig E. Ward
• Program JSAF and JESPP
• MS Operations Support Noble Resolve re HPC,
  Support Install, Test, Training and
  Implementations on new Cluster, Cover all HPCMP
  Reporting and Liaison Requirements
• Ongoing Initiatives Fault Tolerant Comms and
  Computing, Data Management, New Users for J9
  Capabilities

5
Organizational Diagram
6
07 Accomplishments

• Supported HPC and DREN use during Noble Resolve
• FacilitateD the award of new 1,024 core cluster
  to JFCOM
• Represented HPC expertise on weekly phone-con
• Initiated, drafted and submitted five papers at
  I/ITSEC 06
• Presented a required paper at HPCMP UGC 07
• Drove and monitored acceptance testing for new
  cluster
• Initiated and drafted three papers to be
  presented at I/ITSEC 07
• Conceived and prepared to teach GPU programming
  course to JFCOM managers, operators and
  programmers
• Conducted successful JSAF test on new cluster in
  Utah
• Initiated new contacts with IC and DHS
  communities to foster J9 use

7
07- 08 Projected Tasking

• Support HPC and DREN use during events as
  required
• Complete testing and award of new 1,024 core
  cluster
• Represent HPC expertise on weekly phone-con
• Pursue new contacts with IC and DHS communities
  to foster JFCOM/J9 use
• Present three papers at I/ITSEC07 and one at
  WinterSim, all highlighting J9 capabilities
• Present the required paper at HPCMP UGC 08
• Initiate and draft papers to be presented at
  I/ITSEC08 and others
• Teach GPU programming course to JFCOM managers,
  operators and programmers
• Facilitate use and visibility of GPU-enhanced
  cluster at J9

8
Financial Overview

• Financial overview for each task
• List 07 highlights
• Travel
• Personnel and used/projected task hours

9
Contract Overview

• Not direct sub to JFCOM, admin through AFRL
• Currently out of funds
• Reduced staff loadings
• Bob Lucas no charges since Spring 06
• Dan Davis HPCMP PET 1st Qtr, on NSA/DTO now
• Tom Gottschalk Caltech stopped work last
  summer
• Ke-Thia Yao Home Division/NSA/DTO
• Craig Ward Programming for RFQ Editor/ NSA
• John Tran Programming for an Educ. Project
• Gene Wagenbreth NSA/DTO

10
J9 Financial Reporting - Expenditures
Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul
11
J9 Financial Reporting -Travel

• Travel billed to JFCOM this year was to
• Noble Resolve Support at SPAWAR SDiego
• I/ITSEC KYao, and JTran (DDavis and RFLucas
  attended on their own funds authors TGottschalk
  and CWard did not attend)
• HPCMP Cluster Symposium (50 of DDavis travel,
  other half billed to University UNR)
• Travel for next year anticipated to be same, but
  if sufficient funding is not available, some
  conferences will have to go un-attended

12
Labor Hours Against FY07 Funding

• JESPP Agreement is between JFCOM, AFRL (Rome
  Labs) and ISI
• As a rule, academic institutions do not propose
  or account by hourly methods nor do they report
  that way
• All funds were expended on agreed upon tasks on
  behalf of JFCOM and were accounted for as per the
  DCAA-Certified USC Accounting System
• When available funds were expended, professional
  obligations to JFCOM were met by using
  Unrestricted Funds
• No JFCOM effort at ISI was charged to any other
  government project

13
Issues - 1

• Issue Obligations to HPCMP for GFY 08
• Background HPCMP provided 1024 core Linux
  Cluster to JFCOM and will expect
• Visibility of accomplishments with new machine
• Attendance at and submission of paper to
• HPCMP Users Group Conference June 08
• HPCMP Cluster Symposium July 08
• Other conferences of JFCOM choosing (I/ITSEC,
  WinterSim, )
• Showing that Cluster and GPUs were successfully
  employed on behalf of warfighter
• Solution ISI can cover all Liaison with HPCMP
  and all reporting/attendance

14
Issues - 2

• Issue Use of New Cluster and GPU enhanced Nodes
• Background New Cluster is different from
  Koa/Glenn
• - New configuration, 64 bit AMD CPUs, four cores
  per node.
• GPU on each node, i.e. one GTS 8800 per four AMD
  nodes
• Not trivial to implement, but has been shown to
  give 2X performance on some code
• Few organizations currently doing this work
• Implementations not amenable to programmers
  intuitively implementing during operations
• Solution ISI (namely Gene Wagenbreth) is now
  capable of training ops and Guru-ing future
  programming efforts on GPUs and is offering a
  course this fall to initiate that process

15
Issues 3

• Issue Intelligence Community (IC) and Department
  of Homeland Security (DHS) Liaison
• Background IC and DHS need and have shown
  interest in using J9 Capabilities, especially
  with new GPU-enhanced cluster
• Contacts with both IC and DHS have progressed to
  level of seeking
• Statement of capabilities
• Assessment of availability
• Costs of using J9 resources
• Working in concert with J9, progress is imminent
• Solution ISI (principally Bob Lucas and Dan
  Davis) can actively work with these communities
  to ensure J9 capabilities are appreciated and
  employed productively