[linux-lvm] S.A.M.E methodology

[linux-lvm] S.A.M.E methodology

[Ricardo Mattia]

Hello.

	Today we have a host running Oracle 8.1.7 on AIX 4.3.3.
	We decided to migrate this system from RISC to Intel and the
choosed operating system was SUSE Linux.
	Besides that, we are changing the way Oracle Datafile's are
distributed. After some internet research, we got in touch with a
methodology called S.A.M.E, which stands for Stripe and Mirror
Everything. 
	Basicaly, we need to Stripe all the storage available using LVM
(on operating system level) and mirror everything (on storage level). Of
course we increase our risk of an entire system failure in case of just
"one" RAID crash.
	This methodology is explained on Juan Loaiza's whitepaper, from
Oracle Corporation.

	So, my question is : Has anyone implemented this stuff using
Linux SUSE and LVM? If yes, is it work and could you give me some tips
about it? 
	
Thanks in advance.

Best Regards,
Ricardo Mattia
http://www.rge-rs.com.br
51 3218-3168

Re: [linux-lvm] S.A.M.E methodology

[Luca Berra]

On Thu, Jan 08, 2004 at 09:01:13AM -0200, Ricardo Mattia wrote:
>	So, my question is : Has anyone implemented this stuff using
>Linux SUSE and LVM? If yes, is it work and could you give me some tips
>about it? 
>	
i don't know with suse, but i'd use a system that has LVM2 if i wanted
striping and the ability to extend a striped lv.

-- 
Luca Berra -- bluca@comedia.it
        Communication Media & Services S.r.l.
 /"\
 \ /     ASCII RIBBON CAMPAIGN
  X        AGAINST HTML MAIL
 / \

Re: [linux-lvm] S.A.M.E methodology

[Jord Tanner]

This technique, also called RAID 1+0, is often used in database
installations. The basic logic is that with 4 or more drives you can get
the redundancy of RAID 1 with the performance of RAID 0. RAID 5
generally has poor performance for database use. If you have a hardware
RAID card let it do the mirroring (reduces bandwidth on the PCI bus) and
let the OS do the striping. This would require 2 RAID 1 mirror sets,
then striping to those virtual devices. Theoretically in a 4 drive setup
you could tolerate 2 drive failures if the drives are in different
mirror sets. Personally I would use Linux software RAID to create the
RAID 0, then create the LVM volumes on top of that. I'm currently
planning this type of installation, but I've not received the hardware
yet, so I can't tell you my experience.

This has been talked about extensively on the PostgreSQL performance
mailing list. Check out this thread.

Jord Tanner


On Thu, 2004-01-08 at 03:01, Ricardo Mattia wrote:
> Hello.
> 
> 	Today we have a host running Oracle 8.1.7 on AIX 4.3.3.
> 	We decided to migrate this system from RISC to Intel and the
> choosed operating system was SUSE Linux.
> 	Besides that, we are changing the way Oracle Datafile's are
> distributed. After some internet research, we got in touch with a
> methodology called S.A.M.E, which stands for Stripe and Mirror
> Everything. 
> 	Basicaly, we need to Stripe all the storage available using LVM
> (on operating system level) and mirror everything (on storage level). Of
> course we increase our risk of an entire system failure in case of just
> "one" RAID crash.
> 	This methodology is explained on Juan Loaiza's whitepaper, from
> Oracle Corporation.
> 
> 	So, my question is : Has anyone implemented this stuff using
> Linux SUSE and LVM? If yes, is it work and could you give me some tips
> about it? 
> 	
> Thanks in advance.
> 
> Best Regards,
> Ricardo Mattia
> http://www.rge-rs.com.br
> 51 3218-3168
> 
> 
> _______________________________________________
> linux-lvm mailing list
> linux-lvm@sistina.com
> http://lists.sistina.com/mailman/listinfo/linux-lvm
> read the LVM HOW-TO at http://tldp.org/HOWTO/LVM-HOWTO/
-- 
Jord Tanner <jord@indygecko.com>

Re: [linux-lvm] S.A.M.E methodology

[Chris Cox]

Jord Tanner wrote:
> This technique, also called RAID 1+0, is often used in database
...

While both RAID 0+1 and RAID 10 have good performance
characteristics... RAID 10 is much, much more reliable.
Good RAID 10 solutions generally are HW based RAID.
Though I have used Veritas in the past to do a RAID 10.
Not sure exactly how to perform a RAID 10 in software
with Linux.  A single drive failure in 0+1 invalidates
the entire column.  With RAID 10, there's the possibility
of tolerating multiple drive failures.