Opinion required

[Sebastien Masson]

Hi all,

I want to have a survay about who has a Symmetrix storage array.

My symm contain 73GB disk and EMC currently divided them into 10 hyper. I ask EMC to devide my disk into smaller piece so in 32 (the max allowed by physical disk).

In that case I was able to build meta volume and then do a better load balance the symm and prevent any hot hyper.

Do some of you have a higher split then 10 on each physical disk ?

Do you think I'm too crazy with my idea and I'm out of track

Thanks

[Michael Tully]

Hi,

The size of the LUN's being presented depends largely upon what type of servers you plan on connecting.
If you plan on connecting HPUX servers or even a mixture, then yes I would think anything greater than 10Gb LUN's may prove a waste. Remember that if you have multiple volume groups and you start using using huge LUN sizes then you will have a definite waste of space. The balancing of performance can be gained by using EMC powerpath.
If you wish to use NT/windoze servers, then you can create META volumes from the smaller LUN's.

Cheers
Michael

[Vincent Fleming]

It does not really matter what the size of the hypervolumes are... the important thing to do is to present multiple volumes to each host.

For each volume that the host sees, it maintains an I/O queue... multiple queues mean that it can have more I/Os in flight at any given time. This is good for performance.

You also have to think about management. Having a million little devices means having to manage a million little devices (a lot of work!), so you want to find a nice balance where you get several devices per host, yet not have so many as to have a management nightmare.

So, it really depends on how many hosts you have, how much disk space each uses, your performance needs, and your management needs.

If you can tell us (in detail) about your environment (number of servers, OS, storage requirements per server, and performance requirements per server) we can make a recommendation for you.

Good luck,

Vince

[Sebastien Masson]

Ok,

Here more information to help you:

- 5 rp74xx servers on 11i running in a MCSG cluster which run 6 or 7 little db instance. The smalled one can reside into less then 4 GB and the huge one use around 150. All running Oracle 9i.

- All of then are connected trought a switches fabric and use a Symm 8530 as storage system.

- The disk in my Symm are 73GB divide into 9 hyper or 7.56GB.

The problem I have:

Actually I assign for 800GB but everybody use only 280GB, so I have a ratio of 64% of waste space. This is too much for me.

Most of the waste is cause by a small redo, rbs, data, index for all my DB but I can't put everything on one hyper. Those DB are very intensive I/O

I ask EMC to split my device into smaller chunck and use Metadevice to join then in a bigger device with striping.

But EMC told me is not recommanded to split a device in more then 10 for performance issue... But the documentation said you can split a device into a maximum of 32 hyper.. (never sais performance related issue in doc).

:-)

[Vincent Fleming]

Hmm...

7.5GB or so isn't that bad a size, but since you have some smaller DBs, you could benefit greatly from smaller sizes on some of your servers.

Here's some suggestions...

Can EMC make some of the mirror pairs with 32 hypervolumes and some with 10? If so, (I'm not experienced enough with Symm to know this - but you can do this on an XP) then you can have the best of both worlds, larger LUNs for the large DBs, and smaller LUNs for the smaller DBs. If they can't, you can always split the difference, and have them configure 15 hypers per mirror pair... (say, 4-5Gb each?)

You should also be able to make some use of LVM... make several lvols out of each hypervolume when you need less space. Beware that you still need to stripe over several mirror pairs to get performance.

The 4GB database would probably still be the biggest waste of space, though...

Perhaps some other ITRC user can suggest something.

Good luck,

Vince