Scalable supercomputing using COTS (commodity-off-the-shelf)
servers offers the concept of bridging the 20th Century to the 21st and
providing quantum performance/price advantages with the fastest processors,
compilers and interconnects available. Bruce will cover the details
of Compaq's work in scalable supercomputing with reference to the U.S.
DoE ASCI (Accelerated Strategic Computing Initiative) and other peak projects
using Compaq's Alpha processor
with the relevance of these to small, medium
and large implementations in Australia and around the world.
HP MPI Cluster Features
John Cowles, Hewlett Packard, Japan/USA
We discuss the new HP cluster interconnect solution.
This solution provides low latency, high bandwidth, and low cpu consumption
messaging for MPI applications. Asynchronous messaging is supported,
enabling overlapping of computation and communication. The cluster
interconnect can also be used to
perform intrahost transfers, providing datamover
capability in hardware. We present application performance data.
We discuss future plans. Progress and plans
for MPI-2 implementation will be covered. Work with partners including
Platform Computing (LSF),
Argonne National Laboratory (MPI-IO), and Etnus
(TotalView debugger) will be discussed.
Sun Cluster Architecture
Ira Pramanick, Sun Microsystems, Inc., USA
Sun Microsystems is one of the leading vendors
of clustering products, and will continue to make huge investments in clustering.
This talk will present the main features of Sun Cluster, a clustering
solution from Sun Microsystems, that is designed to provide high availability,
scalibility and single-system image. Various components of Sun Cluster
interact together to help achieve these goals, and this talk will describe
three of those components: global networking, cluster file
system and global devices. It will also
touch upon the ease of cluster administration in Sun Cluster.