Titanium: a Java dialect for largescale scientific computing

1
Titanium a Java dialect for large-scale
scientific computing

• Dan Bonachea
• U.C. Berkeley

Titanium Team
Sue Graham Kathy Yelick Paul Hilfinger Phillip
Colella Alex Aiken
Andy Begel Dan Bonachea David Gay Ben Liblit CJ
Lin
Geoff Pike Greg Balls Carleton Miyamoto Peter
McCorquodale Simon Yau
Many others
2
Titanium Overview

• An explicitly parallel, SPMD dialect of Java
• Additional language features for performance and
  support of scientific applications
• Immutable classes (value classes)
• Efficient multi-dimensional arrays
• Region-based memory management
• Parameterized types (templates)
• Operator overloading
• Extensions for high-performance bulk I/O
• Library of useful parallel primitives
• Publicly available distribution
• www.cs.berkeley.edu/Research/Projects/titanium/
• Numerous publications in first-tier conferences
  and journals for both language design and
  applications

3
Titanium Applications

• Heart Simulation Immersed Boundary Method
• First distributed-memory implementation of the
  Peskin-MacQueen algorithm (FFT Navier-Stokes)
• Algorithm also applies to other interesting
  problems
• Microarray oligo-nucleotide selection engine
• Automated selection of good gene sequences from
  a genome, based on various relevant criteria
• Used to create tools for genetic cancer
  research
• AMR Gas Dynamics
• Shock waves propagating through a compressible
  medium
• Various Poisson solvers
• EM charge distribution, gravity fields, etc.

4
Titanium Usability

• Titanium programs run on both shared-memory and
  distributed-memory multiprocessors without change
• Useful for debugging tuning
• Platforms currently supported
• Solaris Linux SMPs/workstations
• Berkeley NOW Millennium clusters
• SGI Origin 2000
• Cray T3E
• IBM SP2 SPPower3 (in progress)
• Users
• Phil Collelas group at LBNL
• ADR group at U. Maryland U. Delaware
• Research group at U.T. Austin
• Parallel computing course at UC Berkeley
• Many students involved grads undergrads