Application Communications and Performance Research

Projects

LAMPI — The Los Alamos Message Passing Interface (LA-MPI) project provides an end-to-end network fault-tolerant message passing system for tera-scale clusters.

Our objective is to deliver to the application developer a production quality open-source software implementation of a message passing library with leading edge technology underpinnings.

The current API supported being MPI 1.2 (C and Fortran bindings).

Research and Development Areas

• Fault tolerance
• Handling transient network failures like data corruption without aborting the application Handling catastrophic failures like the failure of a network component without aborting the application
• Obtaining high aggregate performance over the network
• Scalability to tens of thousands of processors, including application startup and data delivery
• Dynamic process management
• Adaptation to new network hardware and software
• Adaptation to new network hardware and software
• Predictable and efficient data delivery progress

Team leader: Rich Graham, (505) 665-5685, rlgraham@lanl.gov Home page: http://www.ccs.lanl.gov/ccs1/projects/la-mpi/index.html