I've currently got my filers with 'options nfs.tcp.enable off' from when I first set them up due to a perception that this would cause problems.
Has anyone got tcp enabled? Have they seen any problems/benefits, or are there any known issues with this?
We use mainly Solaris 2.5.1/2.7 machines and HPUX 10.20 machines, now with NFS v3 support added, doing compiles and standard (but heavy) file access (no news server, proxy, etc.)
Dave
On Thu, 11 Feb 1999 18:10:03 GMT, Dave Heiland wrote:
I've currently got my filers with 'options nfs.tcp.enable off' from when I first set them up due to a perception that this would cause problems.
Has anyone got tcp enabled? Have they seen any problems/benefits, or are there any known issues with this?
We use mainly Solaris 2.5.1/2.7 machines and HPUX 10.20 machines, now with NFS v3 support added, doing compiles and standard (but heavy) file access (no news server, proxy, etc.)
A note: NFS v3 and TCP/NFS aren't quite the same thing.
We have tcp enabled, but in turn we tell the Solaris boxes to use udp with the "proto=udp" option in the automounter map.
NFS over TCP is a performance hit on a lan compared to NFS over UDP. If you have hosts mounting stuff over a wan link, then TCP/NFS does offer a bit of added stability.
-- Rob ---------------------------------------- Rob Windsor E-Mail - mailto:windsor@adc.com Senior Unix Systems Administrator Voice - phone:972-680-6919 Computer Services Fax - phone:972-680-0370 Business Broadband Group ADC Telecommunications Richardson, TX 75082
Rob Windsor wrote:
NFS over TCP is a performance hit on a lan compared to NFS over UDP. If you have hosts mounting stuff over a wan link, then TCP/NFS does offer a bit of added stability.
I hate to beat a dead horse but...
I don't see why NFS over TCP has to be a performance hit. Aren't there optmizations that could be made that take advantage of the additional semantics that can be gleaned from TCP? ie. since you know where the request came from, isn't there slightly less work to do WRT processing the incoming RPC? I thought I saw someone from Sun (Hal Stern?) or possibly Andy Watson (before he came to NetApp) once say that he saw an implementation that did this.
Or is the problem with the NFS clients?
--tal
Speaking of NFS over TCP, we'd also like to try turning nfs.tcp.enable on to do some testing. Does anyone know if this would affect current mounts if we just enable it at the filer command line with "options nfs.tcp.enable on", or can that be done? Is it a fairly transparent thing or should we schedule down time? Any comments appreciated!
-
Dave Heiland -
Gordon Keegan -
Rob Windsor -
Tom Limoncelli