Andrew Leahy aleahy at
Fri Oct 5 07:43:51 PDT 2001

Steven Timm wrote:
> The problem with making every machine a slave (which we have tried) is
> that eventually the periodic yppusshes themselves load down the net--plus
> you still get timeouts even then.
> Here at Fermilab we have found that NIS starts breaking down when you
> have about 100-150 nodes.  We are moving to a system of shoving out
> password files with rdist..although our situation is easier than
> most because we don't have to push every time someone changes
> a password--these are handled with Kerberos.

I've only been following this thread tangentially, so maybe this has been
brought up before:  How does NSCD change the equation?  Would using this in
conjunction with NIS remove the performance problems, or is NSCD too much of
a burden on client nodes?

Andrew Leahy
aleahy at

More information about the Beowulf mailing list