managing user accounts without NIS
becker at scyld.com
Wed May 24 16:21:04 PDT 2000
On Wed, 24 May 2000, Mark Hahn wrote:
> > Ahhh, we are now using 'bproc' for everything on our clusters, including
> > initial installation and job distribution. It has far less overhead than
> > 'rsh', and putting encryption on the link would make it much slower than it
> do you have any numbers on bproc overhead vs rsh or ssh?
On a 64 node cluster bproc can excute "uptime" on all nodes, including
presenting the output, in about 600msec. There is some execution overlap,
so a single, stand-alone execution takes about 20msec instead of 10msec.
That's a really, really impressive number. Or at least we are impressed;->
We used 'uptime' as our demo because it's a simple, obvious program, and it
shows off the reliability of the cluster.
This execution time includes shipping an arbitrary binary over to the remote
node, rather than loading it from the local disk. For small processes where
the executable is not already in the buffer cache this option is faster than
loading it from the disk. I presume the 'rsh' tests are done with a
preloaded buffer cache?
The 64 node cluster is the old dual-P6-200 128 processor NASA-GSFC cluster
that is now at MTU (Michigan Tech.). That's the same cluster that many of
the large PVFS benchmarks were run on.
Donald Becker becker at scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Annapolis MD 21403
More information about the Beowulf