MD3000 Performance

[Barry Allard]

Hi Peter,

I'd prefer a single LUN > 2 TB as well.  One reason it is wasteful.  One
drive of our 15 x 300 GB setup is unusable because we had to create 2 x 7
disk RAID groups instead of a single, large group because it would have well
exceeded the limit.  (Ignoring the possible RAID group rebuild time issue.)
The other drive is a hot spare.

This seemingly artificial limitation was explained away as a purposeful
feature of the admin utility for windows compatibility.  I'm not buying it.
We were promised a second-hand workaround from the lead software developer
on this product, but it has yet to materialize.  I'd like to ask the Dell
folks: instead of hiding, playing defense, pass-the-buck or tricky
marketing, how about standing behind your product by taking action to
deliver usability to your customers? 

Barry

-----Original Message-----
From: linux-poweredge-bounces at dell.com
[mailto:linux-poweredge-bounces at dell.com] On Behalf Of Peter Kjellstrom
Sent: Friday, July 20, 2007 3:57 AM
To: linux-poweredge at dell.com
Subject: Re: MD3000 Performance

On Thursday 19 July 2007, Andrey Dmitriev wrote:
> The 2TB thing can be worked around with LVM on Linux or ZFS if you
> choose to run Solaris, not sure about Windows (Veritas VM can work on
> all 3 Operating Systems, I amsure)

<RANT>
There is no workaround for a 2TiB lun, it's 2TiB. Using lvm or some other 
volume manager or even a filesystem that uses several block devices does not

work around the core issue, the lun size.

One problem that will remain is that the OS will consider all block devices 
(luns) as completely separate when scheduling IO. This is not true if they 
come from the same underlying raid-set and you can end up with strange a 
performance behavior.

Another problem is introducing another layer (the volume manager). It 
complicates configuration and makes tuning a lot worse (you have to remember

that read-ahead is set on the lv-device but io-sched behavior is set on the 
real device, etc...).

In the end, 2TiB lun max is an embarrassment to the storage industry. 
Individual drives are at 1TB (note, not TiB) today and when they hit 1.25TB 
(soon) then I won't even be able to create a 2 disk raid0 without slicing it

up.
</RANT>

Yes it makes me upset,
 Peter