Serviceguard on raw disksn (without LVM)

[Grunf]

Does anybody know how to configure Serviceguard to run on raw disks (without LVM).
HP-UX 11.31, SGeRAC 10g, EMC disk array.
Thanks

[Stephen Doud]

Serviceguard does not require LVM to create a cluster.
However packages may require disk space to facilitate the application. The standard package control function has the capability activate LVM VGs IF you configure them in the script. If you don't use LVM, your application must handle the storage management aspect.
Also, in a 2-node cluster, split-brain arbitration is required in the form of a quorum server (does NOT require LVM), a lock disk (which requires LVM) or Lock LUN.
So to avoid LVM when implemention arbitration, you will need an external quorum server (either HP-UX or linux-based).

[Grunf]

So the main issue is the quorum server, and the application (SGeRAC???) to handle raw disks?
To resume: configure SG without any disks in ascii config file (that means cmapplyconf will work without VG in it???), install SGeRAC and configure it.
Am i right?
Thanks for the help in advance.

Tomi

[Matti_Kurkela]

ServiceGuard's disk locking is based on VGs. If you use raw disks (/dev/rdsk devices), as far as ServiceGuard is concerned, you're not using disks at all. In this case, ServiceGuard cannot protect you from running the application simultaneously on multiple uncoordinated nodes. You're throwing a way the most important protection ServiceGuard can give you.

You should use "raw LVs" instead of "raw disks". Setup the VG as normal, and when an application requires a device name for raw disk access, give it /dev/vgNAME/rlvolNUMBER instead of /dev/rdsk/cXtYdZ. For the application viewpoint, it's equivalent to a raw disk... but from the sysadmin viewpoint, it is controlled through LVM, with all the LVM's advantages.

Even without ServiceGuard, using LVM with raw LVs allows you to use the VG and/or LV names to give descriptive "labels" to the raw LVs.

For example, without raw LVs:
/dev/rdsk/c4t2d0 = first data disk,
/dev/rdsk/c4t3d0 = second data disk,
/dev/rdsk/c4t5d0 = first index disk,
/dev/rdsk/c4t6d0 = third data disk
etc...

With raw LVs:
/dev/vgdatabase/rlvDATA1 = first data disk,
/dev/vgdatabase/rlvDATA2 = second data disk,
/dev/vgdatabase/rlvINDX1 = first index disk,
/dev/vgdatabase/rlvDATA3 = third data disk
etc...

MK

[Emil Velez]

ServiceGuard 11.18 supports a new functionality called LOCKLUN. See the locklun configuration in the serviceGuard manual for 11.18 version specifically.

This is meant for clusters that use vxvm and no LVM for any disks. The locklun can be a partition on a IA disk or a small lun from a disk array.

[Stephen Doud]

SGeRAC uses either LVM or VxVM/CVM to manage the shared storage. So if you aren't using LVM, I have to assume you are configuring VxVM/CVM-based disk groups and volumes - correct?.

If you prefer LVM, you can permit Serviceguard to designate a cluster-oriented volume group for raw storage as a lock device also, or you may configure a quorum server or LockLUN (A.11.18).

Build the volume groups first, and import them into the second/other nodes.

With LVM-based VGs, the cluster configuration ASCII file must list the VGs using the parameter OPS_VOLUME_GROUP.
The package control scripts should only list the VG names, and not any mount points of any volume groups marked for 'shared' activation.

This information is discussed in detail in the document titled, "Using Serviceguard Extension for RAC", located at http://docs.hp.com/en/ha.html