On Tue, 12 Dec 2000, Rainer Enders wrote:
> Since nfs is not recommended to use what is the
> best solution to setup a cluster solution for a cyrus
> mailserver. I'm thinking of setting up two servers
> that talk to the same ufs mounted filesystem.
You can setup two servers each with their own private volume, but on a
shared storage medium. If one server detects that the other has failed,
it can mount the other's volume, start Cyrus, and take over its IP. You
can use stuff like Sun Cluster for this kind of stuff. You need to write
the switch over scripts though.
> One issue is certainly that the OS filesystem shall
> be a journaling file system since it would take too
> long to fsck a big disk partition. Or you would use
> a commercial product like Veritas that provides its
> own filesystem, with Solaris for example. But that
> is an expensive solution.
Solaris has a native logging filesystem. You just need to turn it on.
> I'm wondering what solutions are people using to
> implement a fault-tolerant system.
Most people working about ensuring single host reliability: redundant
power, ECC memory, RAID, etc.
> Rainer.
Tom
