|
On Fri, 1 Sep 2000, Wayne wrote:
> At 04:28 PM 9/1/00 -0400, Joseph Mack wrote:
> >On Fri, 1 Sep 2000, Wayne wrote:
> >
> >> It will not work for NFS. NFS is working based on server
> >> hand over an opaque packet to the client. Client since
> >> then will communicate with the server based on that opaque
> >> handle. Normal NFS construct that opaque handle involves
> >> some file system ID from that particular server, which most
> >> likely will be different from one server to the other.
> >
> >this is a problem even for UDP NFS? (I know NFS is supposed to be
> >stateless but isn't)
>
> Joe,
> If the NFS servers are identical, there is a chance that it may
> work. However, if the file systems are not identical (from file
> system ID point of view), it will not work, no matter if it is UDP
> or not. The stateless is only true to that particular server. BUT,
> that is the NFSes I had been worked on before based on Sun's
> original invention, it may not true for others implementations.
> Wayne
>
Thanks,
Would the client get a stale file handle on real-server failure?
Now that I think about it, I snooped on a tread with similar
content recently. It would have been linux-ha or beowulf.
I had assumed you could fail out an NFS server, since file servers
like the Auspex boxes use NFS (I thought) and can failover. How they work
at the low level I don't know.
Joe
--
Joseph Mack mack@xxxxxxxxxxx
|
