Bill Pearson (wrp at avery.med.Virginia.EDU) wrote:
: I don't think it is a problem. We have a lot more demand from people
: moving 25 - 100 Meg scanned image files over the network than from
: shared sequence databases.
I admit that NFS _writes_ are much worse in performance that NFS _reads_.
However, we cannot confirm the statement above. We run a in-office subnet
separated from the rest of the world via a cisco box, and find that
2 OSF/1 and 2 IRIX workstations can render the X-performance of Mac and
Interl-based OS/2 and Windows PC useless. If you do the searches in the
background, the _searches_ won't suffer but the interactive work on the
network does. Our net throuput is several terabytes a week. The data rates
produced by a reasonable interface are 2-4 MBit per second, which yields
the expected saturation with already three reasonable workstations.
The original question, whether NFS-sharing data on VMS is a useful thing
to have, can be answered with NO from our UCX perspective. Resources
consumed CPU-wise on the layered NFS are significant. We do a five-digit
number of fasta and blast per year starting from the vax/axp cluster and
are happy that we have the requested setup in place that does not run via
NFS. I have heard people stating that it performs well in the transition
phase, but they never used it heavily in production.
Our use of NFS sharing data is to access transparently read-mounted raw
data for formatting purposes (BLAST, MPsrch) or plain copy to local disks,
which are processes run at night and don't affect interactive use.
R.Doelz Klingelbergstr.70| Tel. x41 61 267 2247 Fax x41 61 267 2078|
Biocomputing CH 4056 Basel| electronic Mail doelz at ubaclu.unibas.ch|
Biozentrum der Universitaet Basel|-------------- Switzerland ---------------|
<a href=http://beta.embnet.unibas.ch/>EMBnet Switzerland:info at ch.embnet.org</a>