becker at scyld.com
Fri Oct 5 08:25:09 PDT 2001
On Fri, 5 Oct 2001, Steven Timm wrote:
> The issue is not only the cpu load on the server, if I understand
> it correctly, but also the network load. 150 nodes doing
> a simultaneous yp lookup is enough to make timeouts.
> >From what I have seen (running 2.2.x kernels and ypbind-1.7-8)
> there is also a yp lookup that happens as long as you have
> files and nis in nsswitch.conf, even if the user is root and
> happens to be found in files. Has anyone else seen this?
Yes. Use 'strace' to watch it happen.
The always-on NIS code is built into glibc, not ypbind. The glibc code
will attempt to communicate with NIS during library initialization, well
before it knows if there are any queries to be made. It's just part of
the glibc bloat.
[[ What was RGB implying when claiming that that NIS "wasn't dark evil"
and then immediately mentioning vampire taps? ]]
It's instructive to study the output of 'strace' during the start-up of
a trivial program. Every I/O attempt has the potential to be a
synchronization bottleneck if the cluster isn't designed carefully.
Those details are important to cluster scalability, which is a separate
issue from application/algorithm scalability.
Donald Becker becker at scyld.com
Scyld Computing Corporation http://www.scyld.com
410 Severn Ave. Suite 210 Second Generation Beowulf Clusters
Annapolis MD 21403 410-990-9993
More information about the Beowulf