The High-Performance Computing Research Group at Coastal Carolina is working on novel techniques to simplify application development and increase performance on modern, large-scale supercomputer systems.
Global-View Programming Models
Modern, large-scale supercomputing systems are comprised of a cluster of individual compute nodes connected with a low-latency switch fabric. Conventional applications written for these systems use message-passing techniques to communicate between parallel processes. Our work focuses on providing an efficient implementation of a global address space that is partitioned over the individual system memories associated with each node. To date, our work has focused on system-level support for linked data structures, such as trees and graphs. Application developers can construct large, distributed tree structures but still write software that feels like thread-like shared-memory programming.
More information on Global-View Programming Models at Coastal Carolina
Orange File System
is a branch of the Parallel Virtual File System
(PVFS) currently under development at Clemson University
. We've been working with the folks at Clemson University to provide some additional functionality related to resilience through redundancy.
More information on Parallel Filesystems at Coastal Carolina
Parallel Programming Education
Coastal Carolina has been recently awarded a mini-cluster from the LittleFe project funded by the Shodor Education Foundation, Intel, Teragrid, and others. This cluster is a 12-core portable unit designed for use in undergraduate computer science curricula for teaching parallel and concurrent computing. The unit supports both shared memory (e.g. OpenMP) and message passing (e.g. MPI) programming styles.
We have received and built our LittleFe unit at the 2011 Supercomputing conference (SC) located in Seattle, WA.
More information on Parallel Programming Education at Coastal Carolina