Robert G. Brown
rgb at phy.duke.edu
Fri Oct 5 07:17:56 PDT 2001
On Fri, 5 Oct 2001, W Bauske wrote:
> Tim Carlson wrote:
> > I am willing to be enlightened as to how my test is flawed. I'll run
> > different tests if asked. Is my test too trivial?
> Add some other net traffic to the NIS server. Basically giving NIS
> the whole bandwidth of your network is unrealistic, at least for things
> I do. For example, if your NIS server happens to also be an NFS file
> server, also start a couple reads/writes to the NFS file systems and
> repeat your experiment and see how that effects your delays.
Another good experiment is to vary the requests or do something that
requires a lot of NIS traffic to facilitate, like creating lots of small
files or reading lots of small files (lots of file stats).
Still, I think Tim's test is reasonable -- NIS performs "decently" up
through maybe 100 clients, especially if there are a few slave servers
to that requests antibunch and clients bind to different servers. The
biggest problems are (again) just what Don and Greg described -- lots of
highly synchronous requests in a true beowulf cluster are bad and NIS
doesn't scale well to 1000's of clients. And the memory leak and other
possible bugs, of course. For ~100 clients or less, for "normal"
poisson-distributed NIS traffic, NIS works adequately. It isn't dark
evil -- it was just designed back when a large network might have twenty
or thirty hosts on 10 Mbps thickwire... (remember vampire taps,
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf