Full HTML for

Basic foilset Remarks on Fox's White Paper and Working Group Report

Given by Geoffrey C. Fox at PetaSoft Foolow-On Meeting San Diego on January 28,29 1997. Foils prepared 2 February 1997
Outside Index Summary of Material

See Software for HPCC Petaflops Architectures - A White Paper For Fox White Paper
Or Computing on the Web New Approaches to Parallel Processing Petaop and Exaop Performance in the Year 2007 for more details on realtion to web Technologies
The first part of these foils summarizes Fox's white paper suggesting need for both fundamental research and careful attention to engineering issues
Use of Web Technologies can help Engineering of Attractive Systems
The second part of talk contains results of a small working group chaired by Fox which met on first day and developed a straw plan to with research and development of "fresh start" software technologies needed for PetaFlop machines
Note Industry view that fresh start was needed -- especially in operating systems.

Table of Contents for full HTML of Remarks on Fox's White Paper and Working Group Report

 Denote Foils where Image Critical
Denote Foils where HTML is sufficient
1 Remarks on Software Issues for PetaFlop Systems January 28-29 1997 workshop San Diego
2 Three General Areas
3 Comments on Fundamental Issues
4 Comments on Data Parallel Languages
5 HPCC-NG Software Infrastructure
6 Working Group Report January 28 1997
7 Proposed PetaSoft Program Structure
8 Components of Initial PetaSoft Program
9 Comments on Operating System Activity
10 Programming Paradigms /Application Activity
11 Programming Environment Tools
12 Role of Technology Transfer to Industry

Outside Index Summary of Material

HTML version of Basic Foils prepared 2 February 1997

Foil 1 Remarks on Software Issues for PetaFlop Systems January 28-29 1997 workshop San Diego

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Geoffrey Fox NPAC Syracuse University
111 College Place
Syracuse NY 13244-4100
HTML version of Basic Foils prepared 2 February 1997

Foil 2 Three General Areas

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Fundamental Issues such as memory hierarchy, latency, data distribution
Engineering Issues such as scaling (exploiting natural parallelism in problems)
Usability Issues such as attractive robust programming environments
HTML version of Basic Foils prepared 2 February 1997

Foil 3 Comments on Fundamental Issues

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Seem to be easiest for PIM architectures for classic geometrically data parallel problems -- even if adaptive and irregular (these are "engineering" issues)
Must extend MPI HPF .. HPJava to support several layers of memory hierarchy
Need to study data movement in key algorithms and how they can be supported in all prototypical architectures
Can possibly use "load balancing ideas" (developed for decomposition onto CPU's today) to address fine grain memory hierarchy decomposition in PetaFlop architectures
Need layered software model with capability to "escape" to lower levels
HTML version of Basic Foils prepared 2 February 1997

Foil 4 Comments on Data Parallel Languages

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Extended message passing will "by definition" allow user to get very high performance
Not clear if penalty of using HPF like approach will be more than classical hope of "factor of two" -- need research and experimentation
  • This is major goal of "fundamental research"
Certainly can move to a "library" like approach (such as Matlab today) with probably interpreted (Java) frontends as an interface to high performance optimized parallel runtime
HTML version of Basic Foils prepared 2 February 1997

Foil 5 HPCC-NG Software Infrastructure

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Need for good Usability and Engineering for a variety of architectures suggest that evolution of today's HPCC software base unlikely to succeed.
Suggest a new Infrastructure HPCC-NG built on Web Technology distributed computing base
Extrapolate from bottom of Computing Cone to the Top!
HTML version of Basic Foils prepared 2 February 1997

Foil 6 Working Group Report January 28 1997

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Geoffrey Fox, Chair
Greg Astfalk
Bill Carlson
Lennart Johnsson
Bob Lucas
Paul Messina
Tom Sterling
Rick Stevens
Marty Sokolski
HTML version of Basic Foils prepared 2 February 1997

Foil 7 Proposed PetaSoft Program Structure

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
We assume near term development focussed on 1-10 Teraflop machines and evolving existing software technology
Current Software Infrastructure (O/S Tools) and concepts (Parallel Environments) will not be able to support petaflop architectures except in isolated cases.
Thus a Fresh start is essential!
Schedule involves point designs to clarify important issues, an ongoing research program and a fresh start developmental phase starting after about 3 years
1
2
3
4
5
6
7
Year
Point Designs
Research
Fresh start PetaSoft Development
Evolutionary Development
HTML version of Basic Foils prepared 2 February 1997

Foil 8 Components of Initial PetaSoft Program

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Scalable Parallel Operating Systems: Essential services include Thread, Memory, Process and I/O Management
  • Neither UNIX nor NT will scale to PetaSystems
  • Study (tighter) Integration of O/S with applications
  • Fault tolerance not so important initially
Programming Paradigms: Expression of Temporal and Spatial Locality in applications so that PetaFlop compiler/Runtime can exploit
  • Including "Gibbs"-like, MPI and HPF Style approaches
Tools: Debuggers, Performance
Problem Solving Environments: Ignore in initial effort
Industry would be potentially interested in such a program!
HTML version of Basic Foils prepared 2 February 1997

Foil 9 Comments on Operating System Activity

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Operating System work must be coupled with real hardware or realistic simulators
  • There are 4000 node systems available now!
A new operating system (suitable for initial use by "first 10-100" users) is a 3-5 year around 10 person development effort
Initial Point Designs would involve looking at O/S issues for particular architectures
  • 2 Industry, 2 Lab, 2 Academia in each of >= 2 teams
  • 6 month studies
HTML version of Basic Foils prepared 2 February 1997

Foil 10 Programming Paradigms /Application Activity

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
Should couple application parallelism study with evaluation of programming paradigms
  • Is there really million way exploitable parallelism in applications
  • Address Evolutionary v Revolutionary approaches
  • Address Implicit (e.g. HPF) versus Explicit (e.g. MPI) approaches
  • The "paradigms" at different levels of abstraction correspond to different "layers" in software
Define around 6-10 interdisciplinary teams with application and compiler/runtime (and of course architecture) expertise
  • ensure full application -- not just kernels -- addressed
  • Possibly will need to change algorithms to enhance data locality etc.
  • However need only look at focussed application(s) in a given study
Point Studies would be followed by research program which develops generally useful programming paradigms integrating results of point studies
HTML version of Basic Foils prepared 2 February 1997

Foil 11 Programming Environment Tools

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
How do performance monitoring and debuggers scale to million way parallelism
Debugging may be easier than some projects to do without real hardware
User Interface challenging!
Not clear that point studies needed -- can start research program immediately
HTML version of Basic Foils prepared 2 February 1997

Foil 12 Role of Technology Transfer to Industry

From Remarks on Fox's White Paper and Working Group Report PetaSoft Foolow-On Meeting San Diego -- January 28,29 1997. *
Full HTML Index
The "fresh start" motivation implies that industry will need a new systems software infrastructure in 15 years
The PetaSoft program must include in initial research/development, a software architecture constraint/study that will enable PetaSoftware to be elaborated into this next generation technical operating environment
Year 1-3: Initial Point Studies/Research
Year 4-6: Develop (with industry involvement) lean mean PetaSoftware to be deployed on 100 Teraflop architectures
Year 7-11: Explore and elaborate PetaSoftware with increasing number of users
Year 11-15: Transfer to industry which develops next generation technical operating system.
Northeast Parallel Architectures Center, Syracuse University, npac@npac.syr.edu

If you have any comments about this server, send e-mail to webmaster@npac.syr.edu.

Page produced by wwwfoil on Sun Apr 11 1999