Los Alamos National LaboratoryProgramming Models Team
Extreme-scale computing specializing in novel computing techniques applied to the hardware-software boundary

Enabling Automatic Parallelization and Transparent Fault Tolerance


We propose the demonstration of a programming paradigm and system specifically to enable the exploration of three mechanisms:

  • implicit (automatic) and explicit task-level parallelism with a guarantee of non-interference,
  • a transparent resilience mechanism less heavy-weight than checkpoint/restart,
  • and explicit speculative task-level parallelism.

The system will have properties that are desirable for scientific computing on highly parallel platforms, such as

  • providing a high level of programming abstraction and being based on well-understood,
  • proven concepts such as the type system and execution model that support both direct programming,
  • and the shallow and deep embedding of domain-specific languages.

The system will also have the properties essential to high-performance computing, such as having performance comparable to C/C++/Fortran and supporting direct linkage to, and interoperability with, programs and libraries written in those languages.