Distributed File Systems

[Bernard Frit]

I'm currently running a 2.0.16 cyrus server mirrored
to another server using :
- 2 external scsi links
- Linux software raid1
- Reiser FS
- heartbeat to manage failover

Thinks are running smoothly excepting :
- the external scsi links require the two servers to be
   very close.
- the external scsi links are very "sensitive" : if someone
   or something shakes the scsi cable there is a serious risk
   to break the link and consequently to break the raid1
   cluster.

The design of this cluster is inspired by IBM's HACMP.

We added an extra feature : cross mirroring.

The two servers are the same :
- NEC P4 1G Ram
- 1 sys scsi HD 9 G
- 2 prod scsi HD 18 G
- 2 mirror scsi HD 18 G
- 4 adaptec scsi cards

Server 1 is running cyrus+postfix+openldap
Server 2 is running apache+mysql+intranet software

Server 1 is mirrored on server 2.
Server 2 is mirrored on server 1.

It's a bit tricky but that way we optimize the hardware
utilization at the price of a degraded backup mode.

The failback procedure is a bit tricky as we have to
reconstruct the raid cluster in the right way.

After 10 month of experience, I wouldn't recommend using
scsi links to do the job but something like drdb to mirror
the volumes via a gigabit ethernet cross link.

--
Bernard FRIT