Обсуждение: Recomendations on raid controllers raid 1+0
Hi chaps,
I'm looking at switching out the perc5i (lsi megaraid) cards from our
Dell 2950s for something else as they're crap at raid 10.
Thing is I'm not entirely sure where to start, we're using 6 SAS
drives and also need a bbu cache. The perc5i has 256mb which I'm sure
would be fine for us.
We're using debian etch 64 bit.
What cards have you lot had success with?
Any tips would be appreciated.
___________________________________________________________
Rise to the challenge for Sport Relief with Yahoo! For Good
http://uk.promotions.yahoo.com/forgood/
Glyn, On Thu, Mar 13, 2008 at 1:33 PM, Glyn Astill <glynastill@yahoo.co.uk> wrote: > I'm looking at switching out the perc5i (lsi megaraid) cards from our > Dell 2950s for something else as they're crap at raid 10. Do you have numbers? Perc 5/i cards perform quite well for us (we have a 8 disks RAID 10 in a 2900 server with the traditional Perc 5/i). -- Guillaume
Bonie++ benchmarks below.
I believe the the Perc 5/i Raid 10 mode is actually a span of mirrors, rather than the expected stripe of mirrors we
shouldexpect from 1+0, and that this is the reason for the shitty performance.
RAID 5
======
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
way5a 32096M 55354 97 201375 59 109586 23 59934 97 427541 33 767.9 1
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
way5a,32096M,55354,97,201375,59,109586,23,59934,97,427541,33,767.9,1,16,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++
RAID 10
=======
Version 1.03 ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
Way5a 32096M 53479 99 131640 33 66718 10 58225 95 339287 25 699.1 1
------Sequential Create------ --------Random Create--------
-Create-- --Read--- -Delete-- -Create-- --Read--- -Delete--
files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP
16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++
Way5a,32096M,53479,99,131640,33,66718,10,58225,95,339287,25,699.1,1,16,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++
----- Original Message ----
> From: Guillaume Smet <guillaume.smet@gmail.com>
> To: Glyn Astill <glynastill@yahoo.co.uk>
> Cc: pgsql-performance@postgresql.org
> Sent: Thursday, 13 March, 2008 3:58:41 PM
> Subject: Re: [PERFORM] Recomendations on raid controllers raid 1+0
>
> Glyn,
>
> On Thu, Mar 13, 2008 at 1:33 PM, Glyn Astill wrote:
> > I'm looking at switching out the perc5i (lsi megaraid) cards from our
> > Dell 2950s for something else as they're crap at raid 10.
>
> Do you have numbers? Perc 5/i cards perform quite well for us (we have
> a 8 disks RAID 10 in a 2900 server with the traditional Perc 5/i).
>
> --
> Guillaume
___________________________________________________________
Rise to the challenge for Sport Relief with Yahoo! For Good
http://uk.promotions.yahoo.com/forgood/
Glyn Astill wrote: > Bonie++ benchmarks below. > > I believe the the Perc 5/i Raid 10 mode is actually a span of mirrors, rather than the expected stripe of mirrors we shouldexpect from 1+0, and that this is the reason for the shitty performance. > Could you build three RAID-1 mirrors on the device, and then stripe that using software RAID-0? Thanks Leigh > > RAID 5 > ====== > Version 1.03 ------Sequential Output------ --Sequential Input- --Random- > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > way5a 32096M 55354 97 201375 59 109586 23 59934 97 427541 33 767.9 1 > ------Sequential Create------ --------Random Create-------- > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > 16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ > way5a,32096M,55354,97,201375,59,109586,23,59934,97,427541,33,767.9,1,16,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++ > > RAID 10 > ======= > Version 1.03 ------Sequential Output------ --Sequential Input- --Random- > -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks-- > Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP > Way5a 32096M 53479 99 131640 33 66718 10 58225 95 339287 25 699.1 1 > ------Sequential Create------ --------Random Create-------- > -Create-- --Read--- -Delete-- -Create-- --Read--- -Delete-- > files /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP /sec %CP > 16 +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ +++++ +++ > Way5a,32096M,53479,99,131640,33,66718,10,58225,95,339287,25,699.1,1,16,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++,+++++,+++ > > > ----- Original Message ---- >> From: Guillaume Smet <guillaume.smet@gmail.com> >> To: Glyn Astill <glynastill@yahoo.co.uk> >> Cc: pgsql-performance@postgresql.org >> Sent: Thursday, 13 March, 2008 3:58:41 PM >> Subject: Re: [PERFORM] Recomendations on raid controllers raid 1+0 >> >> Glyn, >> >> On Thu, Mar 13, 2008 at 1:33 PM, Glyn Astill wrote: >>> I'm looking at switching out the perc5i (lsi megaraid) cards from our >>> Dell 2950s for something else as they're crap at raid 10. >> Do you have numbers? Perc 5/i cards perform quite well for us (we have >> a 8 disks RAID 10 in a 2900 server with the traditional Perc 5/i). >> >> -- >> Guillaume > > > ___________________________________________________________ > Rise to the challenge for Sport Relief with Yahoo! For Good > > http://uk.promotions.yahoo.com/forgood/ > >
On Thu, Mar 13, 2008 at 5:14 PM, Leigh Dyer <leigh@eclinic.com.au> wrote: > Glyn Astill wrote: > > Bonie++ benchmarks below. > > > > I believe the the Perc 5/i Raid 10 mode is actually a span of mirrors, rather than the expected stripe of mirrors weshould expect from 1+0, and that this is the reason for the shitty performance. > > > > Could you build three RAID-1 mirrors on the device, and then stripe that > using software RAID-0? That can be a useful option. You'd have to test it on your setup, but RAID-0 is dead simple, so letting the kernel handle the RAID-0 part is pretty low cost. Especially if you've got lots of CPU power, like many of today's servers.