The question:
>
> The problem is that when they are all booted together we SOMETIMES
> get net hangs or Drive N: (MS-WINDOWS drive) Not Found. If on the
> other hand they are booted at intervals then all is fine.
>
> The problems is that the SPARC 5 can't handle all nfs request
> at the same time, ok that's not surprising. BUT what causes the network
> problems, since I would guess that PCNFS would resend the nfs request
> or does it?
>
> My Questions:
> 1) We have Viglen Ethergen cards and not good old 3COM cards.
> Has anyone else had problems with the above card?
> 2) Is there a problem with PC-NFS?
> 3) I don't believe its SERVER problem since how ever fast the
> server is you are always going to get some collisions!
Answers :
1) Check that there are enough NFS daemons running on the server.
- That's ok I have 20 running but I think increasing them
may help, but still doesn't solve the problem
completely.
2) Someone mention about problems thay had with the license server, but
we have all the licenses local to one machine.
3) Check that the line 'NET PCNFSD *' located in your network.bat
file. This will tell pcnfs to look for a non busy pcnfsd
server to log into.
4) message in whole from one person.
----- Begin Included Message -----
PC-NFS has a configurable retry timeout and count. These fixed values can
always be stretched to their limit if the network has a high enough load.
A not so very neat but workable solution would be to set these values
for indefinite retry. I recall that one of the two (probably count) must
be set to zero through an option in the DEVICE=PCNFS.SYS..... line in
config.sys. (I don't have the manuals within reach, you would have to look
it up yourself). This will fix read failures.
Another dirty hack to help. Mount the N-drive twice. If the first one failes
then the second one will succeed. The chances of missing twice are low. If the
first one did succeed then the second one will give a error messages that can
be safely ignored. This will fix your mount failures.
Also you can specify the mount (net use n:) to retry indefinetly trough options
in the commandline (/timeout=nnn,retry=0).
BTW. You would probably notice that beefing up the server only worsens your
problem. A bunch of PC all reading the same files in sequence will get near
instantaneous response, as the server will automagically cache the files
through it's underlying file system, so there will be hardly any diskio
involved. This implies that you wouldn't need 20 PC's to fill your ethernet
bandwidth. It can be done with a SPARC 5 and 2 PC's.
Real solutions would involve
- increasing bandwidth, eg. by using multiple ethernet interfaces on the sun.
- throttling boot speed,
- this can be done in software.
Let a booting PC check for a "ticket file" in a shared net directory.
If it exists, wait until it's vanished.
If it's not there create it and do the rest of the booting (win startup).
At the end of the booting phase remove the "ticket file".
- I don't know a Viglen but a lot of PC's have a software function to set the
clock rate (like TURBO /ON). This is real bad advise, but you might want to
set TURBO /OFF during booting.
----- End Included Message -----
Thanks for all who replied
Darren Cooper <cooper@awala.csg.cnes.fr>
rloh@asl.com.hk ("Raymond Loh")
ssuryava@cvimail (Sandeep Suryavanshi)
R.J.Hendley@cs.bham.ac.uk
aribi@geo.vu.nl (Arie Bikker)
cowman@novalink.com (Ed Fortin)
******************************************************************************
Andrew Watkins tel: 0171-631 6720
System Group fax: 0171-631 6727
Computer Science Department
Birkbeck College (University of London)
Malet Street
London
WC1E 7HX
e-mail: andrew@dcs.bbk.ac.uk
http://web.dcs.bbk.ac.uk/CS/Staff/andrew/MyHomePage.html
******************************************************************************
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:13 CDT