EMC powerpath

[hpuxhelp]

What is a MetaVolume ? And how does it different than just vg??

[Wilfred Chau_1]

EMC will "slice" or partition a disk into what they call a hyper, say they slice it up into 4 hypers. And a meta is made up of these hypers. And depending on the bin file on the EMC frame.

For example a meta can be made up of 4 hypers.

Yes, this kind of like stripping.

[Vincent Fleming]

The difference between a Metavolume and a vg is that the Metavolume is done by the disk array and the vg is done by HP-UX.

Note that there are performance ramifications to using Metavolumes in some cases. Particularly, if you use only 1 Metavolume in a vg, it hampers the ability of HP-UX to issue several I/O's at a time (I/O concurrency) since there's only 1 "disk" there.

Better performance is gained by giving LVM several "disks" to stripe over. These "disks" can be Metavolumes, however.

So, you need to strike a balance... too many small LUNs are hard to manage (so you would want to add Metavolumes) and too few LUNs to get good performance.

I hope this helps,

Good luck!

Vince

[Sridhar Bhaskarla]

Steven,

Metavolume is what you see it as a disk on your system. Lemme try to make a picture :-)

___ ____ ____ ____hypervol
___ ____ ____ ____
___ ____ ____ ____
___ ____ ____ ____
. . . .
. . . .
___ ____ ____ ____
disk disk disk disk


Each physical disk is devided into sections called hypervolumes.

Now, take one hypervolume from each disk and combine them to form a single "Metavolume". You can either concatenate them or stripe them to form a single meta volume.

These metavolumes are presented to your system as disks and you create volume groups using them.

The number of disks, hypervolume size and the metavolume combination can very. Check with your EMC engineer for details on your EMC.

Powerpath is a product by EMC that will allow you to autoloadbalance the IO on the system. It can also detect link failures and adjust the IO accordingly.

-Sri

[Brian M Rawlings · Accepted solution]

They really do the same thing, one in HW (in the Symm), the other in SW (in LVM).

With LVM, you pull multiple PVs into a VG, and use the accumulated space as one big group of blocks/extents. In order to distribute I/Os evenly and somewhat consecutively over multiple "PVs" (hopefully over multiple spindles), you add striping to your LVs, in a stripe size suited to the number of PVs in your VG. It gives you whatever space you want, and the striping adds performance if you have I/O patterns that have "hot spots".

With metavolumes (Metas), well, let me back up. In a Symmetrix, physical drives are split into groups of tracks called "hypervolumes", or hypers. A 36GB drive would normally be split into four "9GB" (approx) hypers, a 72GB drive into 8 9GB hypers, etc. Hyper size can vary, this is only a typical setup.

A Meta is a collection of hypers within the Symm, much like a VG is a collection of PVs to LVM. Within a Symm, generally using only one hyper per spindle, you create a large group of tracks that are seen by a host as one large volume (PV). The data is striped across many (typically 4 or 8 or more) spindles, so that each sequential block is on the next spindle, and the next, then back to the first, etc.

This gives you essentially the same advantage as a striped VG, but done in HW. The advantage is that you don't spend CPU cycles on LVM stuff, freeing them up to run cron jobs or play doom. The only real disadvantage is that EMC has to set up or tear down a Meta, so you have to deal with them and schedule with them (and, probably, pay them). With LVM, you can do it yourself, possibly with the box up and running, if you have enough data space available to mirror data to somewhere else, then run in that other space while you fiddle with striping the PVs in the LVs, then mirror back to the original PVs, and then unmirror from the temp space... well, I've done it, but it's a lot easier if you can deal with downtime.

Performance-wise, Metavolumes will likely outperform LVM striping by a few percent, since this is what a Symm is good at, and good for. For many of us, the performance delta is small enough that the ability to do it yourself makes up for it, and more.

The time to use Metas, for me, at least, is when you're installing a Symm and setting everything up right the first time. I'd do it then, in a heartbeat. It's the fastest access you can get out of a Symm, for most data patterns. But about the time we've determined that it's needed, we're long into production, and it would be prohibitive to re-impliment the hypers on the Symm... so we go with LVM striping as the slightly less ideal alternative, and get great performance out of LVM that way.

Hope this helps.

--bmr

[Brian M Rawlings]

Sri: anybody that can draw a picture in this format (no text tags, etc) has my utmost admiration. Great illustration.

Vince: good point on the concurrency. Powerpath is needed to gang up the I/Os on all your channels to fully utilize meta performance.

Later... --bmr

[hpuxhelp]

Now You have helped me to understand more about the different between the two vgs. How do I configure PowerPath? According to the eralier thread, when metavolume plus powerpath = become fast and easy to manage...I would like to know to configure a Metavolume ,what tools do it use? GUI interface or commandline?is it easy like configure pvlinks??

[Jakes Louw_1]

Metavolumes, AFAIK, are set up in the EMC bin-file that gets applied to the proprietary EMC OS on the array.

Allocation of the Metavolumes then become similar to a LUN or hyper: the EMC SE will make it available to a redundant pair or Host Bus Adapters, and you will then see the disks on HP-UX with the appropriate Target and LUN addresses.

However, maybe that has changed with the new Symm 6s?