| Title: | FDDI - The Next Generation |
| Moderator: | NETCAD::STEFANI |
| Created: | Thu Apr 27 1989 |
| Last Modified: | Thu Jun 05 1997 |
| Last Successful Update: | Fri Jun 06 1997 |
| Number of topics: | 2259 |
| Total number of notes: | 8590 |
Customer is using NFS on a SUN server attached to FDDI ring in a 900
Hub
The clients are using NFS and are attached on ethernet.
The systems are linked via DEC900EF
The users can connect to the NFS mounted disks, and then move around
using
cd /this/that/theother without problem.
When they try to list files (using ls command) there is a very long
pause (I believe
up to 1-2 minutes - but i have not timed this myself yet) then the
information
comes up okay.
Next time they do the ls command the information comes up immediately.
This precedure can be repeated on a second /third client etc. The first
time there is
always a delay, but if repeated immediately it is quick to respond.
If the users are then idle for some time, the problem/symptoms recur.
IT IS LIKE something is cached and then 'ages out' ...... can anyone
explain
what is happening and why (also how to fix it ! ;-) please?????)
regards&thanx
dave
| T.R | Title | User | Personal Name | Date | Lines |
|---|---|---|---|---|---|
| 1704.1 | NPSS::MDLYONS | Michael D. Lyons DTN 226-6943 | Tue Jun 06 1995 14:34 | 2 | |
...since you're not having much luck here, you might try the SMURF::NFS
notes conference...
| |||||
| 1704.2 | and the answer is.... | DELBOY::THORNTON | Wed Jun 07 1995 07:46 | 32 | |
Thanks Micheal,
good news is however last night i had some luck on-site!!!
I was studying the SUN docs and came across a reference to Solaris 1
not supporting MTU negotiation.....
I decided to lower the MTU on FDDI (SUN Solaris V2 defaults to 4352
octets for MTU). I changed the fddi servers from 4352 to the ethernet
1500 so that there was no need to IP fragment on our 900EF switch.
This cured the symptoms. The Solaris 1 clients on ethernet now access
the fddi Sun Solaris 2 NFS servers at consistently high speed.
My expanation is that the NFS (which uses UDP - unreliable protocol)
application is losing data perhaps due to buffering or bursty network
traffic (?) and this only shows up when the larger fddi frames are
fragmented by our switch. When there is no fragmentation, the
'overload' does not seem to occur and transactions become reliable as
they were originally on the ethernet.
changing the mtu down (which would lessen the efficiancy of fddi
transfers) is not a problem as all transactions are from
ethernet end-systems to the fddi server
hope this helps someone else someday!
regards
davet
:-)
| |||||