Striped Meta Expansion using RAID5 protection

This was much easier on a CLARiiON...

What about the Virtual LUN options within Sym Optimizer. Have you tried that yet?

yep, once you have enough space deleted you will need to create regular BCVs. I would go with 2-way mirror BCVs since you are expanding production metas ..and you know Murphy's law ...unless you are filling lucky that day

What about performance issues with using concatenated versus striped metas? The data we have on our sym is all OLTP type SQL databases that generate a lot of i/o.

i'll be honest i only have ..maybe a dozen windows boxes on my symm and they are very light-load application servers, while all my oracle db servers are running on Linux/AIX/Solaris. I have not see any performance issues with concat metas on Windows, the load is light plus my hyper size is much smaller then yours ..i use 8.4G hypers on 146/300G drives, so when i do build a ~60G meta ..more likely it will end up on many more spindles then if i were to use 30G hypers. I know why you went to big hypers ..because of those gigantic 500G drives and 256 hyper limit per drive, i am curious how you manage these tiers without accidentally grabbing a 146G hyper and a 500G hyper when you are building your metas.

Actually the large hyper size is because the size of the SQL databases we have and the need for metas in the 400-800GB range for storing database data files. Our goal with the 500GB drives was to have as much space for clones as possible and wrap around of that group of disks was not an issue as the clones are used for development and there are no performances requirements for the development environments.

When we first got the DMX we were provided a copy of the bin file in Excel. We have a legend for the bin file that shows the device ranges of the hypers on each group of physical disks. So when creating a meta I know which range of hypers to choose from and update the spreadsheet with the meta information and use it to create my text files for creating the metas with symcli.

Whenever we add disks to the DMX I get an updated copy of the bin file in Excel and merge it with my version so that everything remains current.

So back to the question of meta expansion... When choosing the devices to use as the BCV during an expansion does it need to be of equal performance? I.E. if the source STD is on 15k drives should the bcv be on 15k drives to provide equal performance during the expansion?

I had the same "issue" long time ago (1 year or so). I did however "found out" that it didn't matter what drive types I'd use for the 2way BCV mirrors. I used 73GB15k spindels as well as 146GB10k, 300GB10k and even 500GB7k2. They all did the trick just fine. And what's more: there was almost no difference in speed / time it took to have the data copied back and forth.

But bottomline: you need true 2way BCV mirrors for expandind a meta with data. Please mind this: it only works on metas, you cannot expand a single symdev to becoming a meta with data on it. Only existing metas can be expanded with data still on them.

Like I said before: I've tried (dedicated) disks as fast as 15k, 10k and even the 7k2 ones, but I haven't notived much difference. Theoretically you are correct, but IRL it didn't matter that much at all.

Since the request is to expand a meta, I'd go with an online meta expansion

What I've also noticed is that it takes about 30 minutes or so to expand a 4 way meta to a 5 way meta (about 34GB of data to begin with - expand from 34GB to about 40GB).
If your lun is heavily used this might be 45 minutes.

But that was on my DMX3....

Aran, Rob is right.
Online meta expansion requires a "real" BCV that will be established (via TimeFinder) to protect your data. The request is for a BCV, wherever you want in the box. Using a slower drive for BCV devices will slow down the expansion process .. But will work.

Maybe you are right .. and maybe it's easier on a Cx .. but you have a DMX4 ..
Enjoy your www.clarion.com storage and live happy ..

Message was edited by:
Stefano Del Corno

Thanks guys for all your answers, that cleared it up for me.

And Stefano I am just surprised that something that was so basic and easy on a clariion is a complex process requiring extra resources on a Symmetrix.

And I am enjoying both my DMX and all my clariions

that's why you get paid the big bucks

I had an opportunity to manage another enterprise class array in HDS Tagmastore. From personal experience ..it is SO much easier to manage for newbies compared to DMX. Hitachi provides a nice java interface that you use to create your LUSe(metas), map them and mask them. Definitely much easier to use then ECC, but then ECC is a lot more then just DMX management tools. I guess that's where SMC came from ..simple, fast and effective.

It is definitely a lot more complex, but that is not the issue for me. It has been interesting to me to learn how to manage the DMX and I appreciate the engineering behind the array.

I think that as a Symmetrix newbie I was a little disappointed that some easy to use integrated features I took for granted on a CLARiiON (not the home theater system ) like fast bind, virtual lun, persistent snaps of clones, etc. were not available or difficult to achieve on a Symmetrix.