vxdisk questions - inconsistent device name

[lshu]

Hi,

I am configuring a Veritas FS using a piece of LUN. The device file in "vxdisk list" is c2t0d6, but the device file is /dev/dsk/c2t1d2.
Is there a way to make them identical?

# vxdisk list c2t0d6
Device: c2t0d6
devicetag: c2t0d6
type: auto
info: format=none
flags: online ready private autoconfig invalid
pubpaths: block=/dev/vx/dmp/c2t0d6 char=/dev/vx/rdmp/c2t0d6
Multipathing information:
numpaths: 2
c2t1d2 state=enabled
c4t1d2 state=enabled

thanks,
Lorne

[A. Clay Stephenson]

I think all you are seeing are multiple paths to the same LUN. In any event, you can't create a filesystem in a "piece of LUN"; it's all or nothing unless you put the LUN under the control of a volume manager --- either VxVM or LVM. You can then create a filesystem in a Volume (VxVM) or a Logical Volume (LVM). The format=none indicates that this LUN is under neither at the moment. If you do not put your LUN under a volume manager then you can still create a filesystem but it will occupy the entire LUN (unless the LUN is partitioned).

[lshu]

Thanks for your quick response.

It is whole LUN that I will use to create a FS under VxVM. Sorry for the confusion. The question is how to manipulate the accessname seen in vxdisk. It (c2t0d6) is not consistent with device file /dev/dsk/c2t1d2.

Lorne

[Mridul Shrivastava]

This LUN would be part of an existing disgroup, else you have to create one.

After that you would be creating a volume then filesystem would be created on that colume.. so unix will see only volume, disk and all vxvm will take care..

So you have to build a dg using this disk or add this disk in an existing dg, once that is done create new vol within that dg then newfs can be executed specifying the vol name.

Only primary c2t0d6 to be used while doing all these task... no need to execute the commands for all the paths as vxvm will take about the multipathing.

[A. Clay Stephenson]

Let me try again.
c2t0d6
c2t1d2
c4t1d2
all refer to exactly the same LUN --- just different ways to get there. This is normal and expected for array LUN's and simple disks have only one path. It's much like telling you to go from Memphis to Mobile. There are many different roads but only one Mobile.

Typically, no one path is any better than any other so pick one and go with it when you add the LUN to your new diskgroup. VxVM automatically detects multiple paths to the same LUN. Should you instead decide to put this LUN under the control of LVM, it it actually necessary to specify the paths to a given physical volume.

[Ralph Grothe]

Hi Lorne,

I would assume that you can change such properties of an VxVM object like accessnames with the "vxedit rename" command (please refer to manpage).
But be considerate about possible implications.

[lshu]

Hi all,

You are all right about what it is. I believe it is normal the way it is. The reason I want to do it is because of a cluster environment. The other node in the cluster with the same hardware, uses the path c2t1d2 in a similar way, I just want them to be identical. Does it make sense to you?

I need to study VxFS more before I can evaluate your answers. BTW, can you pass on some experience regarding setting up a MC/ServiceGuard cluster for Oracle using VxFS? I need a active/passive setup.

thanks very much for your time,
Lorne

[A. Clay Stephenson]

It is very, very rare for the disk devices in a SG cluster to have the same name; nor is it necessary. It is only necessary that the volume group (LVM) or diskgroup (VxVM) be the same. How the individual drives that comprise that volume group are named on each node is unimportant. You keep using the term "VXFS"; you need have no concern about vxfs, it is rock-solid. However, you should note that vxfs refers to a filesystem type and a vxfs filesystem can reside in a VxVM volume, an LVM LVOL, or a raw disk. Using SG, you have a choice between LVM or VxVM as a volume manager. Now if you are setting up an Oracle RAC cluster under SG and want to use a clustered file system (CFS) then you must use VxVM. CFS is a special case of VxFS filesystem in that multiple hosts can access
fully cooked files and all appear to simply be a local vxfs filesystem. Normally, mounting a filesystem on multiple hosts is a disaster because hostA doesn't have a clue about hostB's buffer cache. CFS addresses that problem.

[lshu]

thank you for your reply, that helps.

In my case, it is a active/passive (1+1) - no RAC, using vxfs as file system. Any guidelines on how to configure such a MC/SG cluster for Oracle?

thanks,
Lorne

[A. Clay Stephenson]

Since this sounds as though this is your first SG cluster, the best advice I can give you (though you will probably ignore it) is to attend a SG class. There you will learn that SG itself is rather easy; in fact, after a little practice, you can set up a package like Oracle in a few minutes -- it's that easy. The hard part is getting your hardware/network/environment robust enough for SG and identifying and addressing all single points of failure. If you don't do that (especially network redundancy) you can actually wind up with a system that is LESS reliable under SG than standalone.

Unless you have a compelling reason to use VxVM as your volume manager, I would use LVM. It's simpler (though less powerful) and has been tested under SG much, much longer.
I've played with VxVM under SG but all my production systems still use LVM. One of the biggest decisions you will need to make with respect to Oracle packages is whether the Oracle executables are stationary or move with the package along with the data. My preference is moving with package but that's just me. Obviously, if one of your nodes were IA-64 and another were PA-RISC (which is a legal SG configuration) then would would have its own set of Oracle binaries.

[lshu]

I'm thinking about a SG class too. A good point - I am not ignoring it, however it will be after I'm done with this project.

It is going with vxfs for sure. I'm more comfortable with network and the rest. My main concern is a procedure to set up such a cluster.

any comments,

thanks
Lorne