RE: Raid5 Vs Raid0+1 -- Raw Vs Solaris 9 Concurrent Direct IO UFS
- From: "Nelson, Allan" <anelson@xxxxxxxxxxx>
- To: oracle-l@xxxxxxxxxxxxx
- Date: Thu, 26 Aug 2004 08:53:30 -0500
That's good analysis for single block reads. Then we are left with the
case of multiblock reads and there the issue will boil down to the
=66ollowing factors neglecting cache:
1. Is the block size of the mb read sufficiently close to the stripe
size so that the number of blocks being read crosses 2 or more stripe
boundaries. In that case for that single mb read you could involve more
than 2 physical disks in the read.
2. The read advantage then is concentrated in a hopefully small number
of database motivated reads (if you are doing tons of mb reads, maybe dw
environment you could "maximize" the raid 5 adevantage) and file systems
ops on unix where you want to copy or move multi GB's of data.
Overall, it might be possible, after extensively analysing your system
to gain some read speed advantages using raid 5. You just have to make
sure you don't get the write penalty in production. This also assumes
you find some spiritual benefit to obsessing on your disk configuration.
On the whole I find confining raid 5 to application servers and other
boxes that don't do many writes to be about the only place raid 5 really
makes sense.
Allan
-----Original Message-----
=46rom: oracle-l-bounce@xxxxxxxxxxxxx
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Cary Millsap
Sent: Wednesday, August 25, 2004 4:20 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: Raid5 Vs Raid0+1 -- Raw Vs Solaris 9 Concurrent Direct IO
UFS
=3F
A single-block read from a RAID level 5 array will visit only one block
=3D on the array (unless something weird's going on like a block split
across devices, or there's a partial outage going on).
A single-block read from a RAID level 1 array will also visit only one =3D
block (unless there's a block split issue), but the advantage of RAID
level 1 =3D is that a good controller can fetch the block from the less
busy of two =3D disks storing equally valid copies of the block.
There is no "read advantage" of level 5 over level 1. In fact, it's =3D
quite the contrary. First, because of what I said above. Second, because
a =3D RAID level 5 array has inherently greater load to manage than a RAID
level 1 array. If an application generates R 1-block reads/sec, and W
1-block writes/sec, then the two architectures would compare this way:
- Level 1: would have to process (R + W) I/O requests per second
- Level 5: would have to process (R + 4W) I/O requests per second
So, for example, if write calls comprise 50% of your I/O call workload =3D
(that is, W=3D3DR), then this is your situation:
- Level 1: load is 2R I/O requests per second
- Level 5: load is 5R I/O requests per second
That is, the RAID level 5 system will have to process 2.5x more I/Os per
second than the RAID level 1 system. How could the RAID-5 system keep =3D
up=3F Either with a /lot/ of cache ($$$, and Tim's right; any amount of
cache =3D can be overwhelmed by a high enough sustained I/O rate), or by
buying a =3D /lot/ more disks.
...By the time you buy all that stuff, your whole economic motivation =3D
=66or buying RAID level 5 ("it's cheaper, because you don't have to buy as
=3D many
disks...") is out the window.
RAID level 5 is /not/ cheaper, because you /do/ have to buy as many =3D
disks. And cache. And controller software. ...BAARF.
Cary Millsap
Hotsos Enterprises, Ltd.
http://www.hotsos.com
* Nullius in verba *
Upcoming events:
- Performance Diagnosis 101: 9/14 San Francisco, 10/5 Charlotte, 10/26
Toronto
- SQL Optimization 101: 8/16 Minneapolis, 9/20 Hartford, 10/18 New =3D
Orleans
- Hotsos Symposium 2005: March 6-10 Dallas
- Visit www.hotsos.com for schedule details...
-----Original Message-----
=46rom: oracle-l-bounce@xxxxxxxxxxxxx =3D
[mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Nelson, Allan
Sent: Wednesday, August 25, 2004 3:14 PM
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: Raid5 Vs Raid0+1 -- Raw Vs Solaris 9 Concurrent Direct IO =3D
UFS
Raid 5 does get good read performance because for reads more disks get
in on the action. 3 in his example vs essentially 2 for the raid 0+1
stuff.
Allan
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx put
'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
=46AQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
___________________________________________________________________________=
___
This email is intended solely for the person or entity to which it is =
addressed and may contain confidential and/or privileged information. =
Copying, forwarding or distributing this message by persons or entities =
other than the addressee is prohibited. If you have received this email in =
error, please contact the sender immediately and delete the material from =
any computer. This email may have been monitored for policy compliance. =
[021216]
----------------------------------------------------------------
Please see the official ORACLE-L FAQ: http://www.orafaq.com
----------------------------------------------------------------
To unsubscribe send email to: oracle-l-request@xxxxxxxxxxxxx
put 'unsubscribe' in the subject line.
--
Archives are at //www.freelists.org/archives/oracle-l/
FAQ is at //www.freelists.org/help/fom-serve/cache/1.html
-----------------------------------------------------------------
Other related posts:
