On Tue, Oct 11, 2005 at 08:07:49PM +0100, ed wrote: > On Tue, 11 Oct 2005 21:55:30 +1000 > "Rod.. Whitworth" <[EMAIL PROTECTED]> wrote: > > > RAID 1 (or any RAID really) is NOT a backup. It is a high availability > > system. > > High availability does NOT mean never unavailable. > > Hello again Rod, > > I've been looking at ways to make a redundant and load balanced SAN. As > you put it, it's not high reliability, once you get a problem with RAID, > or the box that it's attached to, you can consider the data 'unknown'. > > The best solution that I have seen is, although a bit of overkill, AFS > (Andrew File System). It's kerberos based authentication on a token > basis. Although I have not implemented it I see that it falls short > because the tokens (if used) expire after 10 hours, which might require > a cron job (if that fails does hell break loose?). > > Because it is limited to a single read/write node per volume, I see that > a volume would be required for every directory that might take more than > a few minutes to replicate to the read only nodes to avoid hammering the > read/write node. > > All the other network distributed file systems seem under developed or > unstable. > > FWIW there is something called DRBD which is considered the closest > thing to RAID-0 over a network, it can fail sometimes with flaky results > in testing. I have found it to be troublesom when problems occur during > sync. > > Do you or anyone else know of anything that works better?
DRBD is RAID-1, actually (with n-way replication under development last time I checked). I assume that was just a typo. ;-) I can't say much more. Testing showed that running DRBD is possible and replication does occur, under fairly non-loaded 'lab' conditions and only testing failover in case of manually failing drives. However, I ultimately decided not to pursue DRBD further. I haven't looked at AFS too much, but seem to recall not looking into it further after realizing the Kerberos auth issue you mentioned. Joachim