Good question !
I'm interested in the performance side of the solution, since I can imagine that allocated space is now scattered around the DMX and not "sequential" anymore.

Anyone care to enlighten us ?

I found some interesting documents after a Powerlink search on "virtual provisioning".

Message was edited by:
RRR

Welcome to the forums !

I wish our "lab DMX4" will be operational soon, so we can start testing this new stuff !!!

Rob I think that VP is so "fresh".

The main point isn't if someone is WILLING to enlightnen you ..
The main point is if someone IS ABLE to enlighten you

I have no experiences to share on VP. Hope someone can shade some light on this topic

Yesterday i read some Whitepapers about VP, all this things sound very interesting!
In oracle u can use concat Metas with thin devs on stripes data dev´s!

So you have striped backend eaysy to expand that allocates only the tablespace size
that is really used!
Also ntfs - Vol in Windowas first use only the space the filesystem needs for metadata!

One doc says the thi dev´s are not manged in Microcode - i thought it´s similar to mainframe ckd - if thes are created in 6572 code you get a seperate volume table to managethis dev´s

So where is the information about thin dev´s stored after creation???

A lot of questions and i hope we will soon get more info on this here so maybe we can
use the eature next year !

PS: you need a seperate licence to use VP - hopefully not accountet for the whole capacity of the Box!!

this pool concept is sort of like Celerra pools, how would you manager performance when you have these gigantic pools of disks. Create dedicated pools for dedicated apps ? ..not efficient, so how do you guarantee these many IOPS for application A ..while application B is so bad that's it's causing poor performance for everybody in the pool. I think manageability will be the biggest challenge.

No experience with Virtual Provisioning (you are brave) but I can confirm that current versions of ECC do not support managing it. This is supposed to be one of the key features added in ECC 6.1 due out in the mid-July timeframe.

The recommendations I heard at EMC world from sites that had done some beta testing with VP devices was that concatenated metas and wide striping at the thin device level were the best way to get performance. Basically the theme was the best configuration was to take a lot of devices and create big VP pools so you would get very wide striping.

A few key things to be careful about in this first iteration:

* Optimizer can't move TDAT devices - there is no "rebalance" possible so if you don't start with wide stripes it could get ugly.

* You can NOT reduce Virtual pools today so plan accordingly.

Be very interesting to see some feedback from everyone on this.

exactly ..same concerns here unless there is a magical way to reclaim un-used blocks? I don't see how it would know. I have tested NetAPP NAS unit with VP, so it's nice and dandy that you have a virtually provisioned volume but then somebody writes a bunch of stuff into it and the volume gets expanded, un-needed data gets deleted but volume at this points is at full size, NetAPP is able to shrink volume back to say data + 20% free space. This is a NAS unit bear in bind ..but still.

I agree on the manageability aspect if you have a lot of diverse applications on a frame that need specific guarantees. I don't think VP makes managing this any easier or harder to be honest just a different way to try and spread-out the load. With anything it becomes how granular do you have to go to meet your requirements without making management too difficult. For a single large random IO environment like an Oracle database I think VP could be ideal because of the way it lets you stripe so wide.

The one thing that concerns me with VP for space savings is that a lot of filesystems like to "touch" the entire volumes they are given. NTFS was one example cited (with Windows being the ideal host for VP that is not good) where it tends to allocate to new blocks before recycling old blocks on the filesystem. If over time (what could be short time) you are going to use the entire space anyhow you won't realize any disk savings. I don't think this will work-out the way most people envision and it was pretty clearly stated that using VP should not be done solely to save storage.

That's exactly what I was thinking ! I think VP is a poor man's way of making money fast. No matter what it's impact is on performance.