How many VG's do I need?

[Vassily Gorbounov]

Dear Colleagues,

That's the silly question - how many volume groups do I need and do I benefit from reducing their number (I think that it would be easier to redistribute space between lvols)? What are the reasons for creating more VG's?

About LV number limitation - I need only 16-17 logical volumes totally.

My configuration - I would like to place my application (non system) data on AutoRAID (~100 GB).
I know that I should use "interlaced" hardware paths for primary and seconday controllers. But depending on the number of VG I will need at least 2 or at least 4 LUNs.

AFAIK I will load all disks and all controllers anyway. So the nature of application data doesn't matter - e.g. if I use two groups of applications - nethertheless I would not split them. I mean that although two "disks" in SAR will be more loaded than two others (in 4 LUNs config) in we will use only one application but all "hardware drives" and controllers will be loaded equally.

And one more question - since I will use lvol stripping - should I use stripe size other than 64K, is there a benefit of using smaller stripe size?

Am I right? What are your opinions?

Best regards,
Vassily Gorbounov
Unilever SNG

[Roger Baptiste · Accepted solution]

hi,

<< how many volume groups do I need and do I benefit from reducing their number >>

That decision is based on few factors:
-the size of the disks you are using
-the number of disks which are used.
-the type of application being run on the
system (database? how many instances?
how many user filesystems?
any other applications?)
- what type of disks? (raid array, emc, xp)

<<(I think that it would be easier to redistribute space between lvols)? What are the reasons for creating more VG's?>>

Three reasons which makes me create more VG's is:
the PE size of VG. I use distributed striping on LV's, so prefer a PE size of 2MB. When the PE size is reduced, i would need to have lesser number of PV's in the VG, which puts limitation on the size of VG.
Since most of my systems uses EMC arrays,
i like to create VG's spread across each emc
controller, rather than group them altogether.
It would be a nightmare to debug performance
or hot spot issues at a later time.
Management-> More Vg's have an advantage in a large system. You can group userfilesystems in one VG, one instance in a bunch of VG's, another instance in another bunch of VG's.

But,for small Raid systems, i just bunch them
into smaller number of VG's.

<>
My configuration - I would like to place my application (non system) data on AutoRAID (~100 GB). only one application but all "hardware drives" and controllers will be loaded equally. >>

In your case, i think one or two vg's should
suffice. (other than VG00). 100Gb and 16 volumes seems small enough to not create more VG's. But, remember to give the right vgcreate options.
vgcreate -e -p < > -s <>

<>

It depends on what sort of application you are using. For internal lv striping, 64K should be fine.

No, this was not a silly question ;-) No question is.

HTH
raj

[James R. Ferguson]

Hi Vassily:

This is not a silly question at all. In fact, it shows good foresight.

First, for two reasons, I would limit the number of volume groups. Subjectively, I think a fewer volume groups with correspondingly, more logical volumes makes disk management easier. In fact, you probably will find configurations that are better able to smooth "hot-spots" by configuring proportionally more physical volumes to fewer volume groups than the inverse. This is particularly true for stripping however implemented ("true" stripes, or logical, distributed extents).

Perhaps more importantly, however, is that for every volume group, a kernel data structure exists thereby using memory resources. The 'maxvgs' kernel parameter controls the ceiling, defaults to ten (10), and can be increased to as much as 256.

You have already noted that you are aware of the desirability of splitting your primary and alternate links between controllers to provide both high-availability and load-balancing. I give high weight to this in laying out any configuration.

However you choose to create your disk configuration, give careful consideration to the 'pe_size', 'max_pv' and 'max_pe' parameters when you issue your 'vgcreate'. None of these values can be changed after volume group creation, and you don't want to find that you have been too conservative later if/when you add additional disk. (Disk stripping also adds wrinkles to long-term management, too). Remember that these parameters affect the size of the LVM tables in the Volume Group Reserved Area (VGRA) and must fit into one physical disk extent.

As for the stripe size, this is best determined emperically by you for your application. Choosing a 64K value for the stripe size is fine.

Regards!

...JRF...