The Three FY99 Alliance Technology Imperatives

1
The Three FY99 Alliance Technology Imperatives

• Increase Capability Computing
• Get Back on the High-End Growth Curve at the LES
• Applications on Clusters of DSM and Dedicated
  DSMs
• Applications on NT Supercluster
• Compare NT and Linux
• Build the Grid
• Computational Grid-PACS National Machine Room
• Access Grid-Linking Desktops and ACCESS centers
• Extending the Grid-MREN, Southern X-roads, Great
  Plains
• Create a Post-Web Computing Environment
• Discipline or Team Specific Portals
• A Generic Workbench Infrastructure
• An Object or Distributed Component Software
  Foundation

2
Increase Capability Computing at LES I The
Hardware Shortfall

• The Alliance Has Done an Outstanding Job of
  Managing the Transition of Users From the PSC and
  CTC, and of Improving Its Service to the National
  Community
• The Transition Left the National Scientific
  Community With a Shortage of High-End Computer
  Resources
• The LES Can Solve This Problem Only by Spending
  More Money for Hardware
• We Urge NSF Provide Additional Capital Funding
  for Accelerated Growth

3
Increase Capability Computing at LES II The
Origin Array

• If New Funds Are Available, the Addition of 256
  Origin Nodes Certainly Seems Justified
• The LES Should Increase its Effort in Working
  With SGI and Others to Increase the SSI As
  Rapidly As Possible
• The LES Should Prepare a Contingency Plan in Case
  the SGI Path Is Untenable and Include
  Alternatives to DSM
• Networking and System Software Need to More
  Closely Coupled

4
Increase Capability Computing at LES III The NT
Supercluster

• Additional Emphasis on Cluster Technology Both
  Promises to Provide a Lower Cost Option, and
  Reduces the Risk of a Strategy That Relies on the
  Viability of a Single Vendor
• It is Essential That the NT Cluster be Tested and
  Made Usable As a Production System As Soon As
  Possible
• Transfer of Large MPI Codes Such As Sugars Would
  Be a Leading-edge Approach That Should Be Pursued
• Andrew Chiens Work Should Be Aggressively
  Supported
• Push NT Toward High-performance, Parallel
  Applications
• Perform a Careful Trade-off Analysis Between NT
  Linux

5
Build the Grid

• The Alliance Has Made a Good Start Towards a
  Future National Network-centric Computational
  Infrastructure (the Grid)
• The Grid Has the Potential to Play an Important
  Role in Focusing and Integrating the Alliance
  Efforts
• A Plan Is Sorely Needed for Further Development
  and Ultimate Deployment of the Grid
• Create a Distributed Leading Edge Site
• Job Schedulers Suffer From Insufficient
  Investigation of Alternatives
• Globus is an Excellent Contribution
• Construct a DWDM Network (16 to 80 Channel
  Oc192c) Within the UIUC Campus
• Funding for the PACS G-force Must Be a High
  Priority
• Create an Access Grid
• An Important Feature of the PACI Institutional
  Model Is Its Distributed and Virtual Nature
• QoS Has Not Received Enough Attention

6
Create a Post-web Computing Environment

• The Alliance Will Ultimately Be Measured by the
  Service It Brings to the Scientific Community
• A Central Theme Among the AT Teams Is the
  Development of Workbenches,
• Easily Accessible Toolsets Intended for
  Widespread Research
• And Educational Use Within the Respective
  Scientific Communities

7
Goals, Roadmaps, and Metrics

• Goals and Metrics
• The Alliance Needs a Stronger Overall Focus on
  Goals and Metrics to Measure Its Progress
• Establishing Measurable Goals and Evaluating the
  Program Based on These Goals Is Critical to
  Continued Success
• In the Next Review Should Be a Comparison of
  Accomplishments Versus the Goals Provided in This
  Review
• Roadmaps and Milestones
• Provide a Global Roadmap for All Alliance
  Projects Not Just the High-end Computing
  Resources
• A Roadmap for Grid Research and Development Would
  Improve Its Chances for Success
• Provide Lists of Clear Milestones Not Only for
  Each ET and AT Team, but Also Milestones
  Specifically Focused on ET-AT Interactions

8
Women and Minorities

• The Alliance Must Make a Stronger Commitment to
  Increase Diversity, Particularly in Its
  Leadership and in ET and AT PIs
• The Average Percentage of Women in Computational
  Science Fields Is Around 12
• Every Year The Alliance Should Report on ET AT
  Teams
• s For PIs (All, Women, Minority) PIs
• s and Numbers For Non-PI Research Staff (All,
  Women, Minority)
• Numbers of (All, Women, Minority) Post-docs

9
Alliance Management Priorities

• Top Level Management Must Maintain a Strong Sense
  of Priorities
• It Will Be Critical That the Essential Goals Are
  Strictly Adhered to As a First Priority
• Serving the Scientific Computation Community
• Advancing the Associated Infrastructure
• Be More Diligent in Using the Grid As a Focus of
  Efforts
• Search for Ways to Keep the Alliance Dynamically
  Changing, Phasing in New Projects and Terminating
  Old Ones
• The Alliance Has Identified Specific Areas Where
  They Intend to Improve
• Including Improving Communications,
• Minimizing Bureaucracy
• Increasing Technical Bandwidth Between
  Applications

10
LES Recommendations From Alliance Site Visit
Report

• The Alliance LES Represents the Most Powerful
  Hardware and Provides Leadership for All the
  Partners
• The LESs Plate Is Remarkably Full and It Must Be
  Diligent in Choosing Priorities.
• Should Provide a Capability for University
  Research That is at Least 20-50 Percent of That
  Offered by ASCI
• Should Create a Campus Ultra-performance WDM
  Alpha-Grid
• Balance Need to Provide Excellent Service to the
  National Users and Need to Devote Cycles to
  Testing the Emerging Grid
• One of the Greatest Impacts NCSA Can Have Is to
  Extend, Refine, Integrate, and Disseminate
  Existing Research Codes, Making Them Easier to
  Use, More Robust, and More Widely Available

11
PACS Recommendations From Alliance Site Visit
Report

• The Alliance Is to Be Commended for Holding These
  Partnerships Together During the Past Year
  Despite the Lack of Funding
• Leverage the ET Activities to Make PACS Even More
  Useful
• Funding the PACS G-Force Must Be a High Priority
• Create An Independent Mechanism for Surveying
  User Satisfaction

12
ET Teams Recommendations From Alliance Site
Visit Report

• The ET Effort Is, Perhaps, Too Broad and It Is
  Not Apparent How Projects Are Pruned and Focus
  Maintained
• Integration Among the ET Tools, a Key Grid Goal,
  Is Not Very Apparent

13
AT Teams Recommendations From Alliance Site
Visit Report

• Measure the Impact of the Infrastructure Work of
  Each AT Team on Its Specific User Community
• Develop Specific Metrics Against Which to Measure
  Progress for Development of New Tools
• As the AT Toolsets Mature, an Important Measure
  of Their Success Will Be Their Acceptance and Use
  by the Wider Community

14
EOT Teams Recommendations From Alliance Site
Visit Report

• The Expertise and Track Records of the EOT
  Leadership are Outstanding
• Roscoe Giles, Greg Moses, and Richard Tapia
• It is Both Logical and Desirable for the EOT
  Teams to Take Advantage of Existing Programs
• The Five Focus Areas Defined for EOT Provide Good
  Initial Coverage of Relevant and Related Areas
• Creation and Adoption of Evaluation Metrics