topic Re: Agile Addressing in SG in Operating System - HP-UX

Agile Addressing in SG

Craig Johnson_1 — Mon, 04 Apr 2011 11:31:12 GMT

I don't think this is an issue but my question is:

Does it matter if disks have different agile addresses from node to node within a cluster?

In the past (11.23) the CTD numbers never matched either from node to node, so I don't think this is a problem, but just wanted to be sure.

Re: Agile Addressing in SG

Ismail Azad — Mon, 04 Apr 2011 11:50:37 GMT

Hi craig,

Yes, this is normally something that worries people but is not_an_issue. As you talk in terms of two different kernels on two different nodes, the likelihood of the DSF being the same is not high and yes this is pretty much of one of the features in terms of an "asymmetric cluster". Not only can you have different persistent DSF on two different nodes, you can have the a legacy DSF on one node and a persistent DSF on the other. In turn which leads to the execution of ioscan -m dsf many times...

Anyways in a nutshell, incomplete vgimport execution will import with legacy DSF if not used with appropriate options and yes there is_no_issue regarding this. Talking about the manageablity aspect came the concept of a cDSF. Don't forget to assign points.

Regards
Ismail Azad

Re: Agile Addressing in SG

Doug O'Leary — Mon, 11 Jul 2011 21:59:33 GMT

Hey;

Just to reiterate: no, no issue. This was one of the central themes to another conversation we had here on ITRC regarding cluster wide device special files: http://h30499.www3.hp.com/t5/System-Administration/dev-cdisk-disk/m-p/4755089#M388582

Hope that helps.

Doug

Re: Agile Addressing in SG

Craig Johnson_1 — Mon, 04 Apr 2011 18:08:12 GMT

Thanks you two! I as trying to reassure a colleague but he wanted something from HP. This and a phone call to back line support gave him the answer he needed - agile addresses don't have to be identical from node to node.

You could see that the correct Symmetrix disk was at the agile address, and because they're LVM disks I knew it was right.

Re: Agile Addressing in SG

Emil Velez — Tue, 05 Apr 2011 02:54:02 GMT

in fact if you have a cluster with dissimilar hardware I can almost guarantte that a disk will have a different address from one node than the other.

THe thing to make sure is the wwid. BUt when you import a disk using the VGID or when the VXVM disk group is activated it figures out the local device file for the disk.

In the olden days when I clustered using K boxes and T boxes(im dating myself) we could never have the same device files on different systems.

Re: Agile Addressing in SG

P Arumugavel — Tue, 05 Apr 2011 03:39:47 GMT

Hi Craig,

Most Serviceguard administrators would prefer a cluster that is symmetric (where the disk device files for shared volume groups are identical between nodes) as it is easier to administer. And while it is possible to alter device file names through the procedure below, this concept is quickly becoming a moot point with the introduction of SANâ s and other disk arrays into Serviceguard environments. With SANâ s and certain arrays, it is nearly impossible to create identical disk device file names between nodes.

In a SCSI disk environment, it is possible to alter instance numbers in order to create identical device file names for disks.

Rgds...

Re: Agile Addressing in SG

Viktor Balogh — Tue, 05 Apr 2011 06:44:19 GMT

> In a SCSI disk environment, it is possible to alter instance numbers in order to create identical device file names for disks.

a little more on reorganizing the IO tree:

http://viktorbalogh.net/blog/hp-ux/hp-ux_sysadm/reorganizing-io-tree-general-steps