Distributed/strict allocation

[Tim Medford]

I have a L3000/11.0 server with 2 Fibre Channel cards (a5158a) connected to a Shark array. Everything is working ok, but I'm having some problems configuring the volume group the way I want.

There are two device files which point to the same LUN on the Shark, one file for each Fibre card. c7t0d0 and c9t0d0.

I created the physical volume, the /dev/vgibm directory and the group file. Then created the volume group with this command:

vgcreate -g PVG3 /dev/vgibm /dev/dsk/c7t0d0 /dev/dsk/c9t0d0

I then created the logical volume using this command:

lvcreate -D y -s g -n ora_rbs /dev/vgibm

Then I try to extend the volume using this command:

lvextend -L 256 /dev/vgibm/ora_rbs PVG3

When I attempt the lvextend I get this message:

lvextend: Not enough free physical extents available.
Logical volume "/dev/vgibm/ora_rbs" could not be extended.
Failure possibly caused by PVG-Strict or Distributed allocation policies.

I think the nature of the problem is that I'm trying to do extent-based striping over two device files, that really point to the same device.

What I'm trying to accomplish is load-balancing across both of the Fibre cards. I didn't want to simply have pv-links for fail-over.

I've done this before with an AutoRAID array, however I had 2 different LUNs in the PVG. The difference now is that I'm pointing to a single LUN and trying to force it to load balance between the cards. Does anyone know if this is possible?

Thanks,
Tim

[melvyn burnard · Accepted solution]

This is not possible with LVM

You have an alternate link that is there for redundancy, not load balancing.
VxVM with the fully licensed verison on HP-UX 11i would give you that functionality

[Patrick Wallek]

Melvyn is right. What you are trying to do will not work.

You can't load balance over 2 controllers to the same disk by striping. To do striping you have to be going to different physical disks (or meta volumes).

Unless you have something like EMC PowerPath to load balance, the best you can hope for with your set up is redundency in case the primary controller goes bad.

For your set up I would just do:

# vgcreate /dev/vgibm /dev/dsk/c7t0d0 /dev/dsk/c9t0d0

This will give you your alternate paths.

# lvcreate -L 256 -n ora_rbs /dev/vgibm

To create your 256 MB logical volume.

There's no point in setting this up in a PVG or doing the PVG-strict allocation since there is really just one disk.

[Tim Medford]

Thanks for the information, I suspected this was the case and just wanted to confirm.

I think what I will do is have our mainframe guy change the allocation on the Shark. If he gives me 2x50gb LUNS instead of a single 100gb LUN I should be able to balance them the way I want.

Thanks again.