regading scsi cable

[#1]

hi

we have two DS-20 Alpha servers with 2 CPU, 4GB RAM & 500 MHz Processor and we are trying to establish a SCSI cable VMS cluster setup among these two servers.

We have configured our servers as follows:








We are getting problem in recognizing the storage box. When we connect the Storage box with the DS-20 server using 68-pin SCSI cable, DS-20 is not able to recognize it moreover, after connecting DS-20 to storage box it is not even showing its internal disks. The same problem occurs when we connect two DS-20 Alpha server using 68-pin SCSI cable without using the Storage box both the server are not showing their internal disks.

Can you please help us how can we make DS-20 server to recognize storage box or how one DS-20 Server can show other DS-20 disks also after connecting each other using SCSI Cable. Is there any system changes or environmental variable to be set for the same.

[Willem Grooters]

I'm missing the picture - it's not shown.

Just a guess: from your description, my guess is that you are trying to establish shared SCSI. I struggle with this myself ;)
This requires specific hardware in DS-20. According the documentation, you'll neeed KZPSA SCSI controllers in EACH box - I tried KZPBY-CY but found it not working properly (but that may have be related to firmware).
You will have to do some environmental changes as well. You'll need to use port-allocation for the SCI but that connects to shared SCSI. You can achieve this by running CLUSTER_CONFIG_LAN.COM, one of the options is usage of shared SCSI. Both your DS-20 systems MUST have the same number on this controller. beware that each system must have a different SCSI-ID on this bus: If one system has 7 (default), you MUST specify the other system a lower number. 6 is the nmost obvious one.
It might be that the controller in the storage box has some requirements as well - but that depends on the controller.

Besides this, you'll have to setup each node as a cluster node and disk server. This is a normal requirement. Again, this is best achieved running CLUSTER_CONFIFG[_LAN].COM.

SCSI alone - without a storage box --- I haven't tried that, but AFAIK above still holds.

HTH so far

WG

[Willem Grooters]

Before I forget: you NEED differental SCSI for shared SCSI.

WG

[#4]

please find the image in atatchment


regards
ajaydec

[Richard Brodie_1]

"SCSI alone - without a storage box --- I haven't tried that"

It would be a singularly pointless thing to do, since the SCSI isn't used as a CI. But you probably knew that.

ajaydec: like Willem says, you need to be using a differential controller for shared SCSI. You can't use the one that the internal disks are on.

[#6]

I am using Differential SCSI only. But as soon as I am connection SCSI cable to the server, server is hanging up and when I removed it, its working fine.
Actually I want to boot both the servers from the SCSI disk located in the storage box, for that DS-20 server should be able to recognize the DISK at the triplet point P0>>>, but when I am giving "show dev" over there its not showing any disks.

[Jan van den Ende]

ajaydec,

Look over Willem's answer again:
>>>
beware that each system must have a different SCSI-ID on this bus: If one system has 7 (default), you MUST specify the other system a lower number. 6 is the nmost obvious one.
<<<

You observed:

>>>
But as soon as I am connection SCSI cable to the server, server is hanging up and when I removed it, its working fine.
<<<

This is the clear symptom of NOT having all addresses unique on the SCSI bus.

Change the Host bus address on ONE of the systems, and then be aware that NO other device is on that address nor ANY device address is duplicate).

hth

Proost.

Have one on me.

jpe

[Willem Grooters]

What controller do you use on the storage box, and what type of SCSI controller do you have in the DS-20 ? (these two must be of the same type (KZPBA - contrary to what I stated earlier. The documentation isn't clear at all: "Guidelines for OpenVMS Cluster Configurations", chapter 6
(8.3: http://h71000.www7.hp.com/doc/82FINAL/6318/6318pro_006.html
7.3-2: http://h71000.www7.hp.com/doc/732FINAL/6318/6318PRO.HTML)

states KZPBA as a requirement for SCSI failover, not KZPSA, but that might not be your intention - follow the manual here.

The manual only covers HSZ70 aned HSZ80 controllers in dual, fail-over configuration - so it really is needed to know what controller you're using.

[#9]

Can we use single ended SCSI cable to make a SCSI Cable cluster system.
I tried to use single ended SCSI Cable storage box and connect it to DS-20 Alpha server which also has single ended SCSI port, this time i changed the default pka, pkb and pkc host id (PKA_host_id) value to 4,5,6 on one of the server and on other system it was 7, but then also its not working.
How shud i go about now.

[Steven Schweda]

> Can we use single ended SCSI cable to make
> a SCSI Cable cluster system.[?]

I thought that I read somewhere that you
can't connect two systems and a disk without
exceeding the maximum cable length for a
single-ended (fast/ultra/whatever) SCSI bus.
I thought that that was why only differential
SCSI was supported. No bets, however.

Doesn't the VMScluster Configuration Guide
(or something like that) explain this stuff?

[Hoff]

>>> Can we use single ended SCSI cable to make a SCSI Cable cluster system.<<<<

Within a cluster, SCSI is and can only be a "cluster storage interconnect". A storage-only interconnect.

There is no host-to-host communications over SCSI. Only host-to-disk.

Hosts can share storage access on a SCSI bus, assuming appropriate controllers and cabling.

For clustering, you need a network, and/or CI, and/or DSSI, and/or Memory Channel (MC) or other such "cluster communications interconnect". You can then add multi-host SCSI -- shared SCSI -- into the cluster configuration.

Fibre Channel is similar to SCSI here, in that it too is a storage-only interconnect.

To help troubleshoot a shared-storage (storage-only) interconnect and to determine if it is even feasible here, please indicate which SCSI controllers are present. Only a subset of the SCSI controllers are supported for multihost operations within an OpenVMS Cluster.

Stephen Hoffman
HoffmanLabs LLC

[Steven Schweda]

There seems to be some info in Appendix A at:

http://h71000.www7.hp.com/doc/82FINAL/6318/6318pro_021.html#scsi_techsum

KZPBA-CB (or -CY), the differential card, is
mentioned, not -CA/-CX, the single-ended
card. Also, in reference to a wide bus,
there's this:

Only the KZPBA-CB is supported in a
multihost SCSI OpenVMS Cluster
configuration.

Table A-4 Maximum SCSI Interconnect Distances
(on the next page) makes sense, except for
this entry:

Single ended Ultra 20.5 m

I assume that the author (or proofreader) was
drunk at the time. For a conflicting (and
more reasonable) summary, see, for example,
"Table 1. The SCSI Family Tree" in:

http://www.noncombatant.org/trove/benway-storage-200005.pdf

[Hoff]

The manuals tend to be less current than other materials, such as the SPD or the QuickSpecs, or the support matrix. The manuals don't tend to get opened and updated quite as often as do these other documents, and can thus trail reality when it comes to supported hardware. FWIW.

http://h18000.www1.hp.com/products/quickspecs/12551_div/12551_div.HTML
http://h18002.www1.hp.com/alphaserver/options/asds20/asds20_options.html

[Colin Butcher]

Hello,

First - I do not recommend this for a production system. It's a simple, crude and unsupported cheap way to kludge up a disc subsystem. Your call - it's cheap, but not pretty and not rock-solid reliable.

I have something similar running here in my lab. It's a AS4100 Galaxy box with 2x FWD (note the differential) SCSI adapters with internal termination (1 per partition), different SCSI IDs (7 and 6) and a "Y" cable between them with the single point of the "Y" going into the front of a DWZZB-VA (FWD to Wide SE SCSI adapter) which is in a BA356 shelf with 2 pairs of 9GB discs.

It's very crude, but in the absence of a fibre or SCSI based disc array it's good enough for my limited purposes. I then use HBVS to create a shadowed system disc. I also had to fool around with SCSI port allocation classes to get it to behave sensibly.

So, it can be made to work, but for a production system I'd do it a different way by using an array controller on a shared SCSI bus, or maybe move to using fibrechannel storage if you can - used HSG80s are cheap enough these days.

Cheers, Colin (http://www.xdelta.co.uk).

[Paul Jerrom]

Howdy,
You can have up to three nodes sharing a common SCSI bus in a 'SCSI cluster' (been there, won't be returning!), but they do take some configuring. You can use a KZPSA, KZPBA or other FWD card, but you do need the correct cabling (with Y cables and external terminators) plus you have to take the internal termination jumpers off the card. You can't 'just' stick a SCSI cable between servers and expect it to work. As others have mentionned too, you need (depending on the VMS version) to either set the SCSI bus ID's, or put in placer SCSI cards so the device is seen as identical from both nodes, or use allocation classes on your SCSI buses. So it's not a trivial task. BTW, I seem to remember also that you can use single ended SCSI as long as you use a DWZZA (or DWZZB?), and also you could use a DWZZH (DOC hub) - in theory at least for up to 4 nodes on the same bus.
FWIW I would set up the cluster first, then add the shared storage in afterwards.
Have fun!
PJ