Alpha 1000A device naming and PCI bus enumeration

[Adrian Graham_1]

Morning all,

We have a customer with two AS1000A, both configured identically with the internal disk shelf running off the 16-bit on-board SCSI, tape/CD running on the on-board 8-bit SCSI and a KZPAA in slot 2 running an external tape drive.

Thing is, one machine has the KZPAA as bus A, on-board as bus B and the other machine is the other way round - we can't find any jumper settings or console variables that will make the bus enumeration happen the wrong way round? We're pretty sure the envars are the same but can't easily check the live machine as they won't let us log on; we only know the internal layout because the top lid interlock was defeated ages ago so we could take the side off :)

cheers

A

[Wim Van den Wyngaert]

Compare the setting of conmsole variable bus_probe_algorithm.

Wim

[Ian Miller.]

there was a description of how this was done in an appendix of the VMS V7.1-1H1 release notes, I don't know if this was copied elsewhere.

there is a probe algorithm console variable which can be set to old or new

The version of VMS is also significant.

[Adrian Graham_1]

Hi folks,

bus_probe_algorithm was the first thing I thought of, but since both machines are running VMS 7.2-2 the algorithm should be set to new since the old setting was for <6.2 or 6.1. I've also got an identical machine on the bench here and tried setting it to old and new anyway, no difference.

I notice there's an ISP1020_EDIT command that dumps the config of the on-board SCSI in hex, so I guess we need to see if we can schedule a shutdown to compare outputs. Documentation for ISP1020_EDIT is pretty much non-existant apart from -sd which loads defaults into NVRAM.

cheers

A

[Steven Schweda]

There's also the ultra-secret bit 1 in the
SYSGEN parameter DEVICE_NAMING:

SYS_PARAMETERS

DEVICE_NAMING

(Alpha only) DEVICE_NAMING is a bit mask indicating whether port
allocation classes are used in forming SCSI device names.

Following is the bit definition:

Bit Definition

0 If 1, enable new naming.

1 Must be 0. This bit is reserved for use by HP.

For more information about port allocation classes, see OpenVMS
Cluster Systems.


Setting (Shhh!) bit 1 gets you VMS device
names which match the console device names.
If it's zero, VMS uses a different algorithm
on at least the PK's, and that can lead to
this behavior.

(According to Google v. comp.os.vms, in 2004,
it was "reserved for use by Compaq.")

[Wim Van den Wyngaert]

Difference in firmware version ?

Wim

[Adrian Graham_1]

Firmware version I can't easily check, but it's on the list of things to do.

As for the DEVICE_NAMING parameter, the devices are seen a different way round at the SRM prompt......

[Ian Miller.]

you can check the firmware revison

$ ANALYZE/SYSTEM
SDA> CLUE CONFIG

[Hoff]

As others have commented, this is usually the search order and the bus probe setting and the firmware, but you might simply want to set the port allocation classes on the SCSI buses and move onto the next problem.

I tend to set the port allocation classes on even the simple SCSI configurations, as it makes growing the configurations over time rather easier. (And it also causes the local folks to become accustomed to seeing and using allocation classes, which makes explaining fibre channel all that much easier.) If you do choose this, avoid allocation classes one and two where you can, as these are expected and used by FC; I set these aside for FC.

I will assume that the external SCSI with the tape drive is not multi-host. Tapes on shared SCSI buses don't work all that well, and HP doesn't support same.

Stephen Hoffman
HoffmanLabs

[Adrian Graham_1]

Hi folks,

We sorted it and it brought back mid-to-late memories of this behaviour - what we failed to do was boot the second machine, just saw DKA at the console and got confused.

AS1000s will boot and keep the device names the same as they are at console level, AS1000As will boot as DKB unless the SYSGEN parameter DEVICE_NAMING is set to 3, it doesn't matter what the allocation class is.

I proved this by moving a system disk between an AS1000 and AS1000A :)

Thanks for the clues, they pointed me in the right direction!

cheers

A