Hello everybody...

I'm (trying) to assemble and set up a RH3.0 Cluster, with the shared storage on a Hitachi SAN, accessed via two (for redundancy) Brocade switches...

Server hardware will be (showing only one node):

HP DL380
4 Gb RAM
2 internal 18 Gb U-320 SCSI (RAID-1 hardware) for root volume
2 Emulex LP9002L fiber channel cards

Each Emulex will be connected to a different switch, to achieve multipath connection to the "fabrics" exported by the SAN.

Other hosts connected to the SAN (with Windows 2K and Solaris 8) have no problem at all with multipath access... i.e. they see only ONE disk for each "fabric", my question is:

"How can I enable multipath access for the FC adapter on Linux???".

I know that I can set up multipathing via MDADM, but... this will be a production machine...

mdadm -C /dev/md0 \
--raid-type=multipath \
--raid-devices=2 \
/dev/sdc /dev/sdd

(assuming that the fabric will be seen as both /dev/sdc and /dev/sdd).

I want a flexible management (via LVM) of the shared storage space, for example allocating one VG to each cluster service... so I can try something like that...

1) creating 14 "8e" (linux LVM) partitions on /dev/sdc (they must be seen on /dev/sdd also)
2) a little big script to create 14 PVs...

export ARGUMENTS=" --raid-type=multipath --raid-devices=2 "
for i in 1 2 3 4 5 6 7 8 9 10 11 12 13 14
do
mdadm -C /dev/md${i} $ARGUMENTS /dev/sdc${i} /dev/sdd${i}
pvcreate /dev/md${i}
done

3) create one VG for each cluster service, assigning one (or more) of the previously created PVs
...
and so on...

Anyway, my question is: "there a recommended (via Emulex utility) way to achieve the same result that I can reach with mdadm?????????"

Best regards and thanks to all!

Hi.

I'm not sure, but I think that you cannot do that without md, or a similar product (like PowerPath for EMC² boxes). Solaris have it's own multipath driver, or use the Veritas one (DMP), for Wintel, don't know ...
I'm actually building 2 Linux clusters, with RHEL 2.1 and 3.0, but with my own cluster software (http://open-ha-cluster.sf.net), on EMC² boxes and on "standard" SCSI drives, and have similar problems. On of them is that the load don't look like being shared between the the two /dev/sdxx devices, it just work for fail-over.
If you find something on the subject, please let me know. We may share our experiences ...

Sam.