I saw your post back in April, but at that time I did not have any
facts. Now (as of 1-May-1998) I do, so here goes.
I just finished setting up a 2 node, AS4100 cluster with the following
parts in each of the systems:
1: AS4100 5/533, 2 cpu's, 2GB memory, OVMS 7.1-1H1, 1 UltraScsi
Adapter (for StorageWorks shelf), 1 DE500-BA, 1 Memory Channel, and 1
KZPSA-BA for the EMC connection. The systems are identical (makes life
easy later on). We built a SCSI hub out of a BA-356 shelf in 16 bit
mode, and 3 DWZZB-VA's (2 in from Alphas, 1 out to EMC).
2: EMC Symmetrix 5700 with FWD SCSI Controller ports. The OpenSystems
volumes are set up as 4.3GB for VMS's viewing pleasure.
I put a copy of my FIS disk on an EMC volume and booted the first
machine there. The first time I did this I made a mistake by not
setting up DECnet Phase IV first and by not loading the licenses for
all DEC layered products for both cluster nodes prior to running
CLUSTER_CONFIG.COM. This cost me about 4 hours of work, but I learned
quite a bit on the way.
BTW, the quorum disk is on EMC as well.
After getting the first system root set up correctly, I simply ran
CLUSTER_CONFIG and added the second root to the system disk. I booted
the second node into [sys1] and off we went.
The cluster is behaving just as one would expect, with the exception
that the Memory Channel is a little "tweaky". I have DEC hardware
support working on this. As long as the machines boot at the same
time, the MC is used for the cluster interconnect. If one machine is
dropped, it rejoins after reboot by using the NI. This is a
documented *feature*, but I don't have to like it.
Today, we are building a cluster with 2 AS8400 5/625's and an AS4100
5/533. The 8400's are identical machines, right down to the placement
of the three PCI busses and the cards in them. These machines have 4
KZPSA's each, even though we will only start out with 2 scsi busses
connected to EMC. We built 2 scsi hubs for the two connections, but
in all ways this cluster will look like the 2 node cluster described
above. We should have this cluster built and running by the end of
The biggest potential source of problems is connecting up the machines
to the storage. We got our DEC Field Engineer to reconfig the
machines so that each machine is identical, as far as the PCI busses
and the adapters in them are concerned. My three words of advice are
"LABEL! LABEL! LABEL!". We laid out every cable in advance, we
labeled the scsi hubs, the cables, the PCI busses, the consoles, the
machines, just everything. By being overly careful at this step, you
are almost guaranteed to not have crossed-connected scsi's or memory
I ended up using both node allocation classes and port allocation
classes, even though it is not required. The reason I did this is so
that I can mount the internal StorageWorks devices from either node to
the cluster without device naming conflicts.
In my opinion, doing a SCSI cluster is a pain in the butt, it is very
complex to set up or change, and a little bit edgy when you are a
traditional CI person. However, the EMC disks are FAST and reliable.
The image backup from the UltraScsi FIS disk to an EMC volume ran in
6.5 minutes, with both /verify and /log !.
The gotcha on EMC is that the microcode has to be very recent to make
the volumes work right with OVMS.
Feel free to email me or phone me if you want to get deeper into this.
>Anybody running EMC Symmetrix storage on their VMSclusters?
>I've asked this question before, but received few answers. This
>time, we're getting closer to actually trying the stuff. And it'd
>be nice if I could talk to somebody that's actually used it, in a
>VMScluster, other than EMC employees talking about what they've
>done in their own lab in Hopkinton.
>Since EMC has graciously provided me with just ONE reference ---
>and they gave me his name on the afternoon before he went on
>vacation --- the outcome is predictable: I haven't had my voicemail
>So now I'm searching for references on my own. Please feel free
>to reply to my e-mail address (below), or to the newsgroups as
>you wish (this message has been crossposted to comp.os.vms and
>EMC, ARE YOU LISTENING? OR ARE YOU AFRAID? Sorry to be so
>inflammatory in a public forum (OK, I'm not sorry), but I've
>been asking you for references for several weeks now. Casually
>dropping names of other big corporations is NOT a reference, by
>the way. I want NAMES and PHONE NUMBERS (or EMAIL ADDRESSES) of
>REAL PEOPLE (other than EMC employees). And please don't waste
>my time, nor the reference's time, by directing me to top-level
>executives --- I want to talk to system managers.
>Scott "not speaking for my employer (not DIRECTLY, anyway)" Snadow
>| Scott Snadow, Former Californian. I wasn't born in Texas, |
>| but I got here as fast as I could! |
Clinical Systems Manager