Stanford Streaming Supercomputer (SSS) Winter Quarter 2002-2003 Wrapup Meeting

1
Stanford Streaming Supercomputer (SSS) Winter
Quarter 2002-2003 Wrapup Meeting

• Bill Dally, Computer Systems LaboratoryStanford
  University
• March 11, 2003

2
Year 2 Overview

• Where we are today
• First year goal was met demonstrated feasibility
  on single node
• Feedback from site visit team was very positive
• Potential for a big impact on scientific
  computing
• But still much to do!
• Key FY03 goals
• Get long-term software infrastructure in place
• Select approach, implement baseline Brook to SSS
  compiler
• Multi-node versions that scale
• Language, compiler, simulator
• Tackle hard problems 3-D, Irregular
  neighborhoods/sparse matrix solve
• Language support, numerics support, evaluate on
  simulator
• Refine architecture
• Cluster organization, aspect ratio, register
  organization, memory organization
• Industrial Partner
• Start serious discussions, outreach to build
  support, close partner in 04

3
Some concerns

• Were doing a great job but
• Losing a bit of focus and momentum
• Tooling on the detail
• Need to take a step back and reexamine the big
  picture
• Need to raise our outside profile
• Publish
• Overview paper
• Brook paper
• Generate some more convincing evidence of
  advantages
• Need a control for bandwidth measures
• Update the web page
• Visit the labs

4
Lets review our overall goal

• Exploit capabilities of VLSI to realize
  cost-effective scientific computing.

5
Review What is the SSS Project About?

• Exploit streams to give 100x improvement in
  performance/cost for scientific applications vs.
  cluster supercomputers
• From 100 GFLOPS PCs to TFLOPS single-board
  computers to PFLOPS supercomputers
• Use layered programming system to simplify
  development and tuning of applications
• Stream languages
• Streaming virtual machine
• Demonstrated feasibility of streaming scientific
  computing in year 1
• Refine architecture and programming system in
  year 2
• Demonstrate realistic applications (3D,
  irregular)
• Build usable compiler
• Resolve architecture questions aspect ratio,
  conditional execution, sparse clusters, reg
  organization, memory system, etc
• Build a prototype and demonstrate CITS
  applications in years 3-6
• With industrial and government partners
• Broaden our base of support

6
Industrial Partner Update

• Candidates
• Cray, IBM, Sun, HP, SGI, Intel
• Initial discussion
• Present SSS project and results to date
• Discuss collaboration models
• Identify next steps
• Met with Cray, Sun, and SGI
• They listened politely, but little traction
• Need more convincing evidence
• Need to address programming issue
• Have to provide a path for legacy codes

7
Outreach

• National Labs
• Los Alamos
• Livermore
• Sandia
• Other Government
• NASA
• DARPA
• DoD (Charlie Holland)
• AFOSR
• User communities

8
Software Win 02 Goals

• Brook
• Define carefully the semantics of the operators
• No progress
• Work on views of memory abstraction
• Proposed API will write up for next SW meeting
• Support for partitioning, shared memory, naming,
  fitting into stream abstraction
• Adopting UPC will write up for next SW meeting
• Support for irregular neighborhoods
• Failed to find an application
• Multithreaded version (Christos)
• Have simple model for multi-node written up
• (NEW) Preliminary Brooktran spec
• Concrete Winter goals Ian/Frank
• Review of the language Pat
• Partitioning (UPC)
• Multi-node/Multi-threaded version
• Irregular support w/ application
• PPoPP paper
• MD on BRT

9
Brook Spring 03 Goals

• Refine semantics of operators
• New version of spec
• Implement views of memory API (UPC)
• Find application for irregular structures
• Dijkstra, incomplete LU
• Dynamic structure
• Start switching to new compiler
• Brooktran spec/implementation
• Implemented in Open64
• Concern have lost metacompiler support

10
Software Win 02 Goals

• SVM
• Spec has evolved
• Concensus between MIT, Texas, Stanford, USC
• Implement multinode version
• No progress
• SVM to simulator path
• No progress
• Multi-thread

11
SVM Spring 03 Goals

• Spec is complete and supports SSS
• Revise single-node simulator
• Multi-node simulator (prelim)

12
Software Win 02 Goals (3 of 3)

• Start regular meetings Done
• Compiler
• Decide on flow from Brook-gtSVM-gtSSS Mattan
• Done
• Select base compiler Jayanth
• ORC, Gnu, SUIF, Tendra, others
• Done
• Spike a simple program from Brook-gtSSS
  Mattan/Jayanth
• Started modified front end operating on WHIRL
• Brook to Nvidia
• Optimizations Spring
• Run time
• Write a white paper

13
Compiler Spring 03 Goals

• Complete feasibility study
• Brook to C path
• Parse Brook
• Generate C
• Optimizations
• See Mattans document
• Need to generate SVM code by mid summer
• Parse Brooktran Alan, Fatica, Jayanth
• Kernel scheduler MULADD Das
• SVM to SSS Francois long term need plan

14
Application Win 02 Goals

• StreamFLOFatica
• Base version is complete
• Not running on simulator
• Early start on 3D version partitioning waiting
  on API def
• StreamFEM Barth
• Waiting on spec for partitioning
• 3D arithmetic kernels done
• Tridiagonal in Brook
• StreamMD Eric/student
• Ported GROMACS to the NV30 benchmarks
• Performance dependent on number of registers
• Doesnt work with CG compiler
• Model applications Ron/Frank
• Started
• Look at Sierra, purple benchmarks ppm, sweep3D
  delay

15
Application Spring 03 Goals

• StreamFLOFatica
• Parse Brooktran F to WHIRL Alan, Fatica
• Partitioned version multi-node UPC
• 3D version
• StreamFEM Barth
• Simulate 3D
• Sparse LUD
• Partitioned version
• StreamMD Eric/student
• Hand-tune NV30 assembly code
• GROMACS in Brook
• Model applications Ron/Frank
• C implementations of adaptive structures
• Look at Sierra, purple benchmarks ppm, sweep3D
  delay

16
Architecture Win 02 Goals

• Single-Node Simulator Jung-Ho, Knight
• 64-bit support, MULADD, Scalar Processor
• Not yet
• Multi-Node Simulator Jung-Ho, Abhishek
• Network model
• Multi-node mechanisms
• Not yet
• Point Studies
• Aspect ratio
• SSE vs VLIW
• Planning
• Conditional execution Mattan/Ujval
• Started
• Sparse clusters
• SRF organization Nuwan
• Complete
• Cache alternatives Jung Ho
• Add and store study Jung Ho
• Started

17
Architecture Spring 03 Goals

• Multi-node simulator
• Point Studies
• Aspect ratio TIM
• Conditional execution Mattan/Ujval
• Sparse clusters Delay
• SRF organization Nuwan
• Refine
• Cache alternatives Jung Ho
• Add and store study Jung Ho
• I/O ?
• Iterative operations Francois
• 64-bit delay
• Scalar Processor delay

18
Special Win 02 Goals

• Fix website Pat
• Public and private websites
• Name that computer
• Mississippi
• Axios
• Submit names to Mattan
• Bill, Pat, Bill to choose
• Project Party Mattan Pats house

19
Name Resolution

• From now on, the SSS is called
• Merrimac

20
Winter Quarter Meeting Schedule

• 4/1 Fedkiw Party
• 4/8 Alan, Fatica Brooktran
• 4/15 Kapasi Conditionals
• 4/22 Fatica StreamFLO update
• 4/29 Review Prep
• 5/6 Review Prep
• 5/13 Tim, Tim StreamFEM 3D
• 5/20 Ian, Pat Brook Specification
• 5/27 Mattan Bandwidth Comparison
• 6/3 Jayanth Compiler
• 6/10 Bill Wrapup

21
Papers

• Arch
• Indexable SRFs (Nuwan)
• Streaming Supercomputer Overview (Tim K.)
• Streaming on conventional CPUs (Mattan)
• Conditionals (Ujval)
• Remote Ops (Jung Ho)
• Aspect Ratio (?)
• Data parallel (SSE) vs. ILP (VLIW)
• Software
• Design of Brook (Ian)
• Data parallel programming on graphics HW (Pat)
• Brook to CG
• Compiler
• Apps
• Gromacs
• StreamFEM (Tim2)
• Overview (Bill and Pat)