lkml.org 
[lkml]   [2012]   [Dec]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [patch,v3,repost 00/10] make I/O path allocations more numa-friendly
Date
James Bottomley <James.Bottomley@HansenPartnership.com> writes:

> On Mon, 2012-12-10 at 12:59 -0500, Jeff Moyer wrote:
>> Jeff Moyer <jmoyer@redhat.com> writes:
>>
>> > Hi,
>> >
>> > This patch set makes memory allocations for data structures used in
>> > the I/O path more numa friendly by allocating them from the same numa
>> > node as the storage device. I've only converted a handful of drivers
>> > at this point. My testing is limited by the hardware I have on hand.
>> > Using these patches, I was able to max out the bandwidth of the storage
>> > controller when issuing I/O from any node on my 4 node system. Without
>> > the patch, I/O from nodes remote to the storage device would suffer a
>> > penalty ranging from 6-12%. Given my relatively low-end setup[1], I
>> > wouldn't be surprised if others could show a more significant performance
>> > advantage.
>> >
>> > This is a repost of the last posting. The only changes are additional
>> > reviewed-by/acked-by tags. I think this version is ready for inclusion.
>> > James, would you mind taking a look?
>>
>> James? Do you have any objections to including this for 3.8?
>
> Probably for 3.9 since the 3.8 merge window is upon us.

OK.

> Do we actually have any performance numbers from the big system people?
> That's really a must for this type of work.

I'm working on getting some numbers.

> It's missing acks from the affected drivers; that's not a show stopper
> but it would be better to have them.

Well, we could trim the driver list to those that have ACKs, but the
driver changes are trivial.

Thanks,
Jeff


\
 
 \ /
  Last update: 2012-12-14 17:41    [W:0.107 / U:0.204 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site