Thanks to all who replied.
The general consensus seems to be to use TCP for NFS although performance
may or may not be improved.
It was suggested to use memory channel for the nfs mounts, however, this is
only an ASE, no memory channel is available.
I'll include the other responses:
---------------------------------------------
tcp NFS will improve connection durability, probably resulting in fewer
timeouts, much less vulnerability to noise on the subnet, and won't suffer
near the "stale file handle" problems you get with UDP.
Just my take on it.
I always use tcp for NFS on alpha Tru64
------------------------------------------------------
Gaining performance is relative using this option. On a lightly loaded
network, you will not see a performance gain because the TCP option
acknowledges every packet whereas UDP doesn't acknowledge anything but
completion. On a heavily loaded network, the TCP option sometimes increase
perceived performance of the transfer at hand, because retransmissions are
limited to missing packets as opposed to the entire file being retransmitted.
---------------------------------------------------------
>Original message:
------------------------------------------------------------------------------------------------------------------------------
>Hello Admins,
>
>We're runnning a 2 member, T64 version 4.0f (patch 3), ASE1.6 cluster.
>In order to achieve some load balancing, the application requires that
>data be shared via NFS services between members acrossed a Fast Ethernet
>private network.
>
>We are currently using NFS version3 over UDP. But I have read that we
>could gain some performance by using TCP.
>NFS over TCP is supported in this version of ASE, but I was wondering if
>there were any known issues to using TCP. Will it really improve performance?
>
>I understand that all that is required to use TCP is use the -proto=tcp
>in the nfs mount command.
>
>Any comments would be appreciated.
>
>Thanks.
>
>John
Received on Mon Feb 25 2002 - 14:42:08 NZDT