PVG

[anand_51]

hi,
what is PVG and why we are using this?

Regards,
Anand

[Torsten.]

Hi,

this mechanism is useful if you want to mirror with more than 2 disks in a volume group, see

man lvmpvg

http://docs.hp.com/en/B2355-60130/lvmpvg.4.html

[Matti_Kurkela]

PVG = Physical Volume Group.

Basically, it's a way of dividing PVs of one VG into PVGroups and then applying certain LVM policies in a per-PVG basis instead of per-PV basis as normal.

The LVM can use the PVG information to limit where mirror copies of each extent can be placed. If you don't use PVG, LVM will assume that all the PVs in a VG are equal.

For example, when you have four disks but just two SCSI controllers (2 disks to each controller), you'd prefer to have one of the mirror copies on controller A and and the other on controller B.

If both were on controller A, a cabling problem or a hardware fault could disable both mirrors at once, crashing the system.

Separating the mirror halves on different controllers may improve performance too: any operation can be run in parallel on both SCSI buses, instead of requiring one bus to transfer the data twice.

Anything the PVGs can do, you can also do manually if you're careful: the PVGs just allow you to automate some of the basics.

[Mridul Shrivastava]

We use PVG when we want to use striping with Mirroring so we keep all the striped PVs in a PVG and then mirror one PVG with another...

Once we create a PVG vg sees it as a complete LV .. no matters how many disks are their in PVG..

To set up this we need minimum 4 disks so each PVg will have 2 disks and u can stripe them and then they can be mirrored across another PVG..

This kind of setup helps us in getting more I/O.. so speed is higher and at the same time we have a mirror as well..

[MHudec]

PVG is just a bunch of physical volumes in volume group with something in common (like connection via same hba, or like origin from same disk array).

Some examples might be:
1.) high availability
We have a disk array (XP, EVA, you name it) in two datacentres (dc A, dc B) and we want to mirror the volumes like original stay in dc A (these physical volumes are grouped together in PVGdcA) and mirror stay in dc B (these physical volumes are grouped together in PVGdcB). So when dc A gets nuked, we still have our data available from dc B.

2.) io distribution
We have few disks/physical volumes (name them pvolA, pvolB, pvolC) from disk array and we want to distribute io load between those physical volumes for performance reasons. So we create PVG from these physical volumes and create logical volume with it's extents distributed to physical volumes in that PVG via round robin (see -D y and -s g options for lvcreate command). Like then first logical extent is mapped to physical extent on pvolA, second logical extent is mapped to physical extent on pvolB, third is mapped to pvolC, fourth is mapped (again) to pvolA, ... like here:

pvolA = c13t2d1
pvolB = c13t2d2
pvolC = c13t2d3

LE PV PE
0000 /dev/dsk/c13t2d1 0000
0001 /dev/dsk/c13t2d2 0000
0002 /dev/dsk/c13t2d3 0000
0003 /dev/dsk/c13t2d1 0001
0004 /dev/dsk/c13t2d2 0001
0005 /dev/dsk/c13t2d3 0001

Hope this shed some light on the issue.

[anand_51]

thank u for ur valuable answers