HPC (High Performance Computing)
"A supercomputer is a computer 1 order of magnitude less powerful
than is needed to solve the problems we have right now."
Problems with other definitions.
. . . The supercomputer of today is a personal computer tomorrow
. . . Real measure is the ability to solving "hard" problems
. . . Clock speed isn't the most interesting or important parameter
. . . Backplane/bus bandwidth, cache size, memory access speed, disk bandwidth
SMP (Symmetric MultiProcessor) and clusters/farms
. . . Most appropriate solution is problem dependent
. . . Bandwidth limitation of clusters defines their applicability
Parallel processing for clusters
. . . rsh, PVM, MPI, sockets
. . . Trivially parallel algorithms
. . . "True" parallel algorithms, "Sum of the parts is greater than the whole"
. . . Comparison between ECDL and SETI
Issues as a cluster grows above 16, 32, 64, 128 machines
. . . Reliability - Time Between Failures (TBF)
. . . Bandwidth to shared disk
. . . Network topology
. . . Beowulf Cluster
. . . Clusters of SMP machines
. . . World wide virtual computing
. . . Swinburne cluster of Dec Alpha processors