SCSI Attached Drives In Blade Centre Unusable

[Robert Atkinson]

I have a cluster of 4 blades, each with a SFF SAS drive. I want to use these for Page and Swap files, but each disk is seen as DKA200.

I've switched off MSCP, but each node in the cluster still sees the other nodes disks, even though they're physically difference drives :-

SYSMAN> do mc sysgen show mscp_load
%SYSMAN-I-OUTPUT, command execution on node ALPHA
MSCP_LOAD 0 0 0 16384 Coded-valu
%SYSMAN-I-OUTPUT, command execution on node BETA
MSCP_LOAD 0 0 0 16384 Coded-valu
%SYSMAN-I-OUTPUT, command execution on node GAMMA
MSCP_LOAD 0 0 0 16384 Coded-valu
%SYSMAN-I-OUTPUT, command execution on node SIGMA
MSCP_LOAD 0 0 0 16384 Coded-valu
SYSMAN> do sh dev dk
%SYSMAN-I-OUTPUT, command execution on node ALPHA
Device Device Error Volume Free Trans Mnt
Name Status Count Label Blocks Count Cnt
$1$DKA200: (ALPHA) Mounted 0 ALPHA_SWAP 286746624 1 1
%SYSMAN-I-OUTPUT, command execution on node BETA
Device Device Error Volume Free Trans Mnt
Name Status Count Label Blocks Count Cnt
$1$DKA200: (BETA) Mounted 0 (remote mount) 1
%SYSMAN-I-OUTPUT, command execution on node GAMMA
Device Device Error Volume Free Trans Mnt
Name Status Count Label Blocks Count Cnt
$1$DKA200: (GAMMA) Mounted 0 (remote mount) 1
%SYSMAN-I-OUTPUT, command execution on node SIGMA
Device Device Error Volume Free Trans Mnt
Name Status Count Label Blocks Count Cnt
$1$DKA200: (SIGMA) Mounted 0 (remote mount) 1
SYSMAN>


Is there any way I can change the SCSI ID's or stop the cluster seeing each others drives?

Rob.

[Ian Miller. · Accepted solution]

why not set a port allocation class of 0 for PKA on each system then the device name of each disk will become nodename$DKA200

[Joseph Huber_1]

Strange You have mscp_load == 0, but disks are served.
Show us MSCP_SERVE_ALL.
If it is not == 0, probably mscp_load is implicit (confirmed?).

And in addition all cluster members have ALLOCLASS=1: all scsi disks from all systems are served as $1$DK.

To not cluster-serve local disks, set MSCP_SERVE_ALL=0,
and a different ALLOCLAS per system,
or alloclass=0 on all systems, so the disks appear as nodename$DKA200.
or change the SCSI-ID to be unique on each blade.

[Robert Atkinson]

Ian, is there a way of setting the PKA allocation class without going through CLUSTER_CONFIG?

Rob.

[Robert Atkinson]

Don't worry - I thought I had to run through the whole of the ClusConfig script, but there's a direct menu option for the allocation class.

Cheers, Rob.

[Hoff]

Everybody that's first working with blades with OpenVMS tends to hit some variation of this case.

Please lob a software enhancement request at HP; this is a common configuration problem with the BladeSystem configurations, and CLUSTER_CONFIG or (less desirably) the documentation could do a better job of not landing you in the middle of this case.

[The Brit]

Hi Rob.

I want to forward a suggestion that Hoff gave me.

Set the ALLOCLASS parameter to a different value on each node in your cluster, (I use 10, 20, 30, 40, etc.).

I keep a bootable backup of my system disk on one of the internal drives (very useful for making clean backups, etc.)

however
If your System Disk is a shadowed SAN disk, then,

if ALLOCLASS is set to 0, and you back this up to an internal disk, you will get a BUGCHECK if you try to boot that internal disk.

The reason is that Shadowing internal disks requires ALLOCLASS be non-zero. This being the case, the boot will be unable to create the shadowed system disk using the internal disk.

You have two options at this point.

1. as discussed above, set ALLOCLASS to a non-zero value. or,
2. Boot conversational and set SHADOW_SYS_DISK to 0.

Just my 2 cents worth.

Dave

[Robert Atkinson]

Ran CLUSTER_CONFIG on each node and configured allocation class for PKA SCSI adapter to be unique to each node.

Rob.