[ Thanks to Moshe Bar
for this link. ]
“The Linux world is in clustering frenzy. A scant few months
ago, the only solution was to write your own clustering software
for high availability. True, for the scientific-clustering arena,
there was also the option of using Beowulf or message-passing
clustering libraries such as PVM. However, apart from these early
solutions, Linux lagged behind proprietary and commercial
clustering options.”
“Things look decidedly more bright now. We have SGI’s Failsafe
almost completely ported, we have the Linux High Availability
project, we have the Piranha cluster by Red Hat, and many
others.”
“There are also some very viable scientific clustering
solutions. The best of which is Mosix. Mosix implements a Single
System Image cluster (SSI). SSI clusters are best because they give
users the impression of working on a single giant computer instead
of a cluster of individual nodes. Users can log onto any node
of a Mosix cluster and launch processes that run anywhere in the
cluster deemed appropriate by the clustering software in the
kernel. If users correctly implement DNS they just telnet to
cluster.mynet.com and the DNS round-robin feature lets them log
onto a random node and fork programs. And fork and fork and fork.
In fact, the Mosix technology is often described as
“fork-and-forget.”Some Mosix clusters count among the top 50
supercomputers in the world. Having enough unused computing
equipment in my lab I decided to build a sizable Mosix cluster
myself. I took the following hardware and installed the Mosix
kernel on all of them….”