Hi,

My client is getting a new SAN solution. It will use VRAID5 and will
have
seemingly boundless storage cabability. I am happy enough with the
storage aspects.

The bit I am worried about is the Virtual RAID from a performance
point
of view. There is not much info out there on it.

- How do the usual RAID10 vs. RAID5 decisions vis a vis the read and
write
performance dilemma on a "normal" box stack up in a VRAID
environment?
- Are Whalen's articles on RAID configuration in his book on
Performance
Tuning now irrelevant in a VRAID context?

What happens when all the database applications are sitting on the one
SAN.
Surely there is going to be a bottleneck somewhere.

It is really strange how little VRAID chatter exists on the Web... Is
this so new that nobody really knows. Or maybe it really does solve
all storage and performance problems ;-)

Regards

Liam Caffrey

Hi

The parity calculation will still be an overhead, so there is no getting
away from that.

The same rule also apples to aligment. Aligning the OS block start and size
with the volume's block start and size is important too, and possibly more
important as VRAID it uses little blocks off the disks and not the whole disk.

We looked at it and decided for simplicity of managment and standardization,
to stay with RAID-10.

Regards
Mike

"LCaffrey" wrote:

Although mainly Oracle based the reasons are the same see BAARF (Battle
Against Any Raid Five) http://www.miracleas.dk/BAARF/BAARF2.html

http://www.miracleas.dk/BAARF/RAID5_versus_RAID10.txt

GertD (AT) SQLDev (DOT) Net

Please reply only to the newsgroups.
This posting is provided "AS IS" with no warranties, and confers no rights.
You assume all risk for your use.
Copyright © SQLDev.Net 1991-2004 All rights reserved.

"LCaffrey" <lxc (AT) eosys (DOT) ie> wrote

Mike,

HP have a thing called RAID5 ADG (Advanced Data Guard) which consumes 6 I/Os
for each logical write. Is there any chance they have incorporated the same
into VRAID5

block start and size? Say I wanted to carve up 8 volumes for assigning to the
various logical aspects of the production database: logs, tempdb, 3 x indexes,
etc... would there be a separate OS block start for each volumne block start.
Could you point me towards some material on this please.

Regards

Liam

Hi

"Block Alignment" is when the drive RAID blocks and the OS formatted block
start at the same place.

Say you have a RAID-5 drive set configured with 64kb blocks over 5 drives.
When you create this though the RAID card software, it sets up 5x64kb blocks,
one block on each drive. This is repeated until the whole drive is
initialized.

Now when you format the drive from an OS perspective, say with 64Kb blocks,
you don't want the OS 's first 64 kb block to start anywhere else but
exactly "aligned" with the RAID blocks. i.e. you don't want 1/2 the first OS
block to be in one RAID block and the other half on another RAID block
(because if this occurs, one OS read requires 2 hardware reads)

You need to make sure that the RAID blocks vs OS block are sub-divisible in
each other, preferable, the same.

Regards
Mike

"LCaffrey" wrote:

Mike,

That is very clear, thank you.
How can I check if this is the case or not, i.e. to make sure that the
1st OS block starts at exactly the same place as the 1st RAID block
and that both block sizes are identical (and, I guess, equal in
number)

It would be rather unfortunate if they were misaligned ;-)

Regards

Liam

"Mike Epprecht (SQL MVP)" <mike (AT) epprecht (DOT) net> wrote