The question was:
Solaris 2.5 includes NFS V3 which is supposed to provide
asynchronous write capability. Until now we have used PrestoServe
to achieve this.
Does Solaris2.5/NFS V3 effectively obsolete PrestoServe?
The Answer is:
No, NFS V3 DOES NOT OBSOLETE PRESTOSERVE.
File Integrity is not assured in case of a crash.
Below are the 3 responses:
>From blymn@awadi.com.au Thu Jan 11 22:16 CST 1996
From: blymn@awadi.com.au (Brett Lymn)
I don't think so - prestoserve provides ansync nfs by caching the data
to battery backed RAM. Just enabling asynch writes on NFS would mean
that your data would be lost if the machine crashes before the write
actually happened, with prestoserve the writes would be done after the
machine came back up.
-- Brett Lymn, Computer Systems Administrator, AWA Defence Industries =============================================================================== >From Jean-Claude.Giese@loria.fr Fri Jan 12 05:09 CST 1996 From: Jean-Claude Giese <Jean-Claude.Giese@loria.fr> I think yes but not completely. I've read that the gain in write with PrestoServe was under 10% with NFS v3. (it's ~ 80% with NFS v2).I think it will be useful so long that they are still clients in NFS v2 (not in SunOS 5.5).
Yes we have PrestoServe cards in all ours NFS servers :-( and we plan to migrate in Solaris 2.5.
>From rogerk@veritas.com Fri Jan 12 10:44 CST 1996 From: "Roger B.A. Klorese" <rogerk@veritas.com>
The V2 NFS protocol required synchronous operation because of NFS = statelessness. If there were no such requirement, a client, for = instance, could do a write which did not hit the disk on the server, = have the server crash and reboot, and do another I/O after that reboot, = and assume the data would be consistent.
V3 allows, by option, synchronous or asynchronous NFS.
Prestoserve is designed to provide *synchronous* NFS without the latency = caused by disk speeds. Even if a server crashes, data is guaranteed to = reach the disk eventually.
Asynchronous NFS makes no such guarantees. It's an *option*. You can = say, "I trust my server not to crash, so I'll forego the safety net of = synchronous I/O."
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:10:51 CDT