philipp@gris.informatik.uni-tuebingen.de (Philipp) (05/31/91)
Hello,
We have got the problem that SGI and HP workstations don't like to work
together via NFS. Here is the problem:
We have a large network (university) with several SGI-workstations on the
backbone net.
Attached to it is a HP discless cluster (5 HP400t at one server) that
is connected to the backbone by a UNIX-PC (386 running Interactive Unix) that
is only acting as a router (no other tasks).
Connections from either the HP's or the SGI's to other hosts on the net (via
the router or not) are no problem, but access between SGI and HP (either
direction) don't work. The problem is mostly apparent on large files, when
a programm accessing it, halts with the message 'NFS server not responding'.
After a while the message 'NFS server OK' appears, but (on the next package ?)
we start again with 'not respondig'.
Small files seem to be ok, most of the time.
Maybe others have had this problem before an can send suggestions on a fix.
Any help is welcome.
Here the data on the machines:
HP: Serie 9000, HP400t, HP-UX 7.05 (7.03 is same problem), 16MB, 3x660MB SCSI
disks on the server.
SGI: W-4D210GTXB and W-4D25TG-I, IRIX 3.3.2, 16MB, 300MB local disk each
Network:
The backbone is a thick ethernet, while the cluster is on a thinwire.
.
. (lots of other machines) .
. . ( 5 more HP's)
| +----------+ .
+---------------| SGI | |
B | +----------+ | +---------------+
a | +-----| HP Client |
c | +----------+ | +---------------+
k +---------------| SGI II | |
b | +----------+ | +---------------+
o | +-----| HP Server |
n | +----------+ | +---------------+
e +---------------| UNIX PC |------------+
| +----------+
.
.
Philipp Slusallek Universitaet Tuebingen, WSI/GRIS
Auf der Morgenstelle 10-C9, D-7400 Tuebingen, FRG
Email: philipp@gris.informatik.uni-tuebingen.de Phone: +49-7071-295462
--
Philipp Slusallek Universitaet Tuebingen, WSI/GRIS
Auf der Morgenstelle 10-C9, D-7400 Tuebingen, FRG
Email: philipp@gris.informatik.uni-tuebingen.de Phone: +49-7071-295462mg@ (Mike Gigante) (06/01/91)
The problem may be the PC router - we had a similar problem at this site (b/n two SGI machines) The PC ethernet cards have a small (4k? ) buffer, anyhow smaller than the default packet size in NFS. You can change the NFS packet size to be smaller that the PC buffer and then things worked for us. see the nfs man poage for details. Mike Gigante, RMIT Advanced Computer Graphics Centre
jason@hpcndjdz.CND.HP.COM (Jason Zions) (06/03/91)
Sounds like you're overrunning the router. Try using -o "rsize=4096,wsize=4096" in the mount command; this will reduce the size of the packet train in requests and responses. You can go smaller on the sizes if you like, but the sizes must be powers of two, and you'll gain no benefit in this configuration from going below 1024. -- This is not an official statement of The Hewlett-Packard Company. No warranty is expressed or implied. The information included herein is not to be construed as a committment on HP's part. The devil made me do it. This won't save me from the lawyers' wrath, but it can't hurt. Jason Zions The Hewlett-Packard Company Colorado Networks Division 3404 E. Harmony Road Mail Stop 102 Ft. Collins, CO 80525 USA jason@cnd.hp.com (303) 229-3800
raj@hpindwa.cup.hp.com (Rick Jones) (06/03/91)
Check your RPC retransmission rates (/usr/etc/nfsstat) on either side of the router - they are probably high - perhaps a result of overrunning the PC router's buffering capacity. You can try decreasing the mount size to something like 1K or so. It will 'slow-down' NFS, and hence make life easier on your PC router. rick jones ___ _ ___ |__) /_\ | Richard Anders Jones | HP-UX Networking Performance | \_/ \_/ Hewlett-Packard Co. | "It's so fast, that _______" ;-) ------------------------------------------------------------------------ Being an employee of a Standards Company, all Standard Disclaimers Apply