Port allocation class for CDs

[Ziggy Filek]

Some time ago I successfully set up port allocation classes for local SCSI disks on my 3xES80 Fiber Channel cluster. In this case device name one uses is NOT the disk drive device name("DKA"), but the controller device name ("PKA","PKB",...), resulting in the following (or similar)entries in the sys$system:sys$devices.dat:
...
[Port $PKA]
Allocation Class = nn
...
What should I use in place of the "PKA" in the example above if the device in question is a CD-rom drive (DQA0). Do I simply use "DQA" or there is a controller involved here, and I should use some other name?

Thanks for any input.

Ziggy

[Robert Brooks_1]

Port allocation classes are for SCSI devices only. DQ devices are IDE; you cannot set a PAC for an IDE device. If your sysgen param ALLOCLASS is non-zero, your IDE devices will be prefixed by that allocation class, rather than SCSNODE.

In general, PACs are only required for devices that may be directly accessed by more than one node. Since an IDE bus is not shareable, PACs are not needed.

-- Rob

[Jess Goodman · Accepted solution]

Yes, you can just use DQA and/or DQB.

Note that depending on the VMS version you are running you might have to manually edit SYS$SYSTEM:SYS$DEVICES.DAT. IIRC CLUSTER_CONFIG.COM does not support these devices, at least through VMS 7.3.

[Jess Goodman]

Robert,

As you said: (non-zero) "PACs are only required for devices that may be directly accessed by more than one node". And "If your sysgen param ALLOCLASS is non-zero, your IDE devices will be prefixed by that allocation class". But there in lies the problem.

If you have two or more nodes with these devices and the same non-zero system ALLOCLASS value then you will have a device name conflict and it will prevent you from MSCP serving them. Setting the PAC to zero for these devices is the work-around.

I use PACs on all nodes that have a non-zero system allocation class so I can set the DQ/DV and any other unshared SCSI bus device names to be node$DQA0, node$DVA0:, etc. Even if there wouldn't be a device name conflict in the cluster I think this makes it more clear that these are local devices.

[Robert Brooks_1]

Yes, you can just use DQA and/or DQB.
----

Touche! I stand (or sit) corrected, having just confirmed that on one of my systems.


-- Rob

[Ziggy Filek]

On all 3 nodes of my cluster the ALLOCLASS system parameter was originally set to 1, presumably because VMS docs refering to HSV-based Fiber Channel SAN (EVA 5000) specify that allocation class has to be 1 for fiber channel. As far as I know, there is no way to associate allocation class with the HSV110 controller, on the controller itself. The "Guidelines for Cluster Configuration" manual states (somewhat enigmatically) that VMS "forces" ALLOCLASS=1 for Fibre Channel attached disk devices and ALLOCLASS=2 for Fibre Channel attached tape devices. Just out of curiosity (since my current configuration works fine): If I defined different alloclass on each node (say 4,5 and 6) with DEVICE_NAMING set to disable port allocation classes, would this still work, i.e. woud VMS still "force" alloclass to be 1 on every DGAnn:, 2 on every MGAnn: and would attach appropriate allocation class to local devices like $4$dka0, $5$DKA0, $4$DQA0 etc. ?
Do you have experience with such a configuration in a Fibre Channel environment?
This would be an alternative to use port allocation classes in SAN environments.

Ziggy

[Jess Goodman]

Yes, fibre channel disk devices are always set to be $1$DGAn irregardless of systen pararmeters ALLOCLASS, DEVICE_NAMING, or the SYS$DEVICES.DAT file.

[Bill Hall]

Ziggy,

The following urls point to TIMA documents that explain device naming and setting port allocation class on VMS v7.1 and above:

http://h18000.www1.hp.com/support/asktima/operating_systems/CHAMP_SRC981117004322.html
http://h18000.www1.hp.com/support/asktima/operating_systems/CHAMP_SRC000330002406.html

Bill

[Ziggy Filek]

Thanks to everybody for your help. I now have enough info to perform my task.

Ziggy