becker at scyld.com
Thu Oct 4 17:32:20 PDT 2001
On Thu, 4 Oct 2001, Tim Carlson wrote:
> On Thu, 4 Oct 2001, Greg Lindahl wrote:
> > BTW, by slaves, do you mean "slave servers" or "clients"? There's a
> > big difference. Having lots of slave servers means a push takes a
> > while, but queries are uniformly fast.
> I meant clients.
> 1 master, 50 clients.
> The environment on the Sun side wasn't a cluster. 50 desktops.
Completely different cases.
Workstation clients send a few requests to the NIS server at random times.
Cluster nodes will send a bunch of queries simultaneously.
> Never had complaints about authentication delays. I just haven't seen
> these huge NIS problems that everybody complains about.
The problems are not failures, just dropped and delayed responses. A
user might not notice an occasional ten second delay. When even trivial
cluster jobs took ten seconds, you'll notice.
> If you were running
> 1000 small jobs in a couple of minutes I could imagine having problems
> authenticating against any non-local mechanism.
Hmmm, a reasonable goal is running a small cluster-wide job every
second. I suspect the NIS delays alone take longer than one second with
just a few nodes.
> Our current cluster builds use http://rocks.npaci.edu/ for clustering
> software. This system uses NIS. I know it is odd to hear of any other
> system than Scyld on this list, but we have had good luck with NPACI
We don't discourage discussions about other _Beowulf_ systems on this
list. We have thought extensively about the technical challenges
building and running clusters, and are more than willing to share our
experiences and solutions.
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