Linux Archive

Linux Archive (http://www.linux-archive.org/)
-   Ubuntu Server Development (http://www.linux-archive.org/ubuntu-server-development/)
-   -   software raid and multiple cores (http://www.linux-archive.org/ubuntu-server-development/602999-software-raid-multiple-cores.html)

Mark van Harmelen 11-26-2011 08:35 AM

software raid and multiple cores
 
Hi
Thanks to those who contributed to a discussion on software raid recently, it changed my mind about the universal desirability of*hardware*raid cards.*
So now I'm intent on building a sw raid based*machine, but mostly am interested in performance under disc load.*

Basic question: given a number of spare cores (be they hyperthreaded or not) is ubuntu's sofware raid clever enough to be able to deal with multiple read and write requests simultaneously, one request per spare core? Or would I expect only to see one of the spare core utilised?

thanksmark*
--
ubuntu-server mailing list
ubuntu-server@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-server
More info: https://wiki.ubuntu.com/ServerTeam

"Peter M. Petrakis" 11-28-2011 02:07 PM

software raid and multiple cores
 
On 11/26/2011 04:35 AM, Mark van Harmelen wrote:
> Hi
>
> Thanks to those who contributed to a discussion on software raid
> recently, it changed my mind about the universal desirability of
> hardware raid cards.
>
> So now I'm intent on building a sw raid based machine, but mostly am
> interested in performance under disc load.
>
> Basic question: given a number of spare cores (be they hyperthreaded
> or not) is ubuntu's sofware raid clever enough to be able to deal
> with multiple read and write requests simultaneously, one request per
> spare core? Or would I expect only to see one of the spare core
> utilised?

Request level might be the the wrong level to look at, Linux already has
lots of queue level optimizations. What's really going to matter here is
efficiently handling parity calculations and the effective RAID
real estate (stripes). There does appear to be a multi-core optimization
for MD RAID5/6 and it's disabled by default on Ubuntu kernels.

http://lxr.linux.no/linux+v3.1.3/drivers/md/Kconfig#L157

config MULTICORE_RAID456
bool "RAID-4/RAID-5/RAID-6 Multicore processing (EXPERIMENTAL)"
depends on MD_RAID456
depends on SMP
depends on EXPERIMENTAL
---help---
Enable the raid456 module to dispatch per-stripe raid operations to a
thread pool.

If unsure, say N.

$ grep MULTICORE_RAID456 /boot/config-3.0.0-1*
/boot/config-3.0.0-12-generic:# CONFIG_MULTICORE_RAID456 is not set
/boot/config-3.0.0-13-generic:# CONFIG_MULTICORE_RAID456 is not set

It looks like a work in progress, I hope you have nothing to lose on those disks.
See the linux-raid list for more information: http://marc.info/?l=linux-raid .

Peter

>
> thanks mark
>
>

--
ubuntu-server mailing list
ubuntu-server@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-server
More info: https://wiki.ubuntu.com/ServerTeam

Mark van Harmelen 11-28-2011 04:18 PM

software raid and multiple cores
 
Thanks Peter,
Your description of my question was spot on :) thanks.Luckily we dont have those discs set up yet... I shall enquire further on linux-raid as to what experimental might actually mean in this context.

regardsmark

On Mon, Nov 28, 2011 at 3:07 PM, Peter M. Petrakis <peter.petrakis@canonical.com> wrote:





On 11/26/2011 04:35 AM, Mark van Harmelen wrote:

> Hi

>

> Thanks to those who contributed to a discussion on software raid

> recently, it changed my mind about the universal desirability of

> hardware raid cards.

>

> So now I'm intent on building a sw raid based machine, but mostly am

> interested in performance under disc load.

>

> Basic question: given a number of spare cores (be they hyperthreaded

> or not) is ubuntu's sofware raid clever enough to be able to deal

> with multiple read and write requests simultaneously, one request per

> spare core? Or would I expect only to see one of the spare core

> utilised?



Request level might be the the wrong level to look at, Linux already has

lots of queue level optimizations. What's really going to matter here is

efficiently handling parity calculations and the effective RAID

real estate (stripes). There does appear to be a *multi-core optimization

for MD RAID5/6 and it's disabled by default on Ubuntu kernels.



http://lxr.linux.no/linux+v3.1.3/drivers/md/Kconfig#L157



config MULTICORE_RAID456

* * * *bool "RAID-4/RAID-5/RAID-6 Multicore processing (EXPERIMENTAL)"

* * * *depends on MD_RAID456

* * * *depends on SMP

* * * *depends on EXPERIMENTAL

* * * *---help---

* * * * *Enable the raid456 module to dispatch per-stripe raid operations to a

* * * * *thread pool.



* * * * *If unsure, say N.



$ grep MULTICORE_RAID456 /boot/config-3.0.0-1*

/boot/config-3.0.0-12-generic:# CONFIG_MULTICORE_RAID456 is not set

/boot/config-3.0.0-13-generic:# CONFIG_MULTICORE_RAID456 is not set



It looks like a work in progress, I hope you have nothing to lose on those disks.

See the linux-raid list for more information: http://marc.info/?l=linux-raid .



Peter



>

> thanks mark

>

>



--

ubuntu-server mailing list

ubuntu-server@lists.ubuntu.com

https://lists.ubuntu.com/mailman/listinfo/ubuntu-server

More info: https://wiki.ubuntu.com/ServerTeam



--
ubuntu-server mailing list
ubuntu-server@lists.ubuntu.com
https://lists.ubuntu.com/mailman/listinfo/ubuntu-server
More info: https://wiki.ubuntu.com/ServerTeam


All times are GMT. The time now is 07:45 AM.

VBulletin, Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO ©2007, Crawlability, Inc.