public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
* BIO ordering and NativeCommandQueueing
@ 2004-06-15 10:02 Guillaume Lacôte
  2004-06-15 11:36 ` Jens Axboe
  0 siblings, 1 reply; 2+ messages in thread
From: Guillaume Lacôte @ 2004-06-15 10:02 UTC (permalink / raw)
  To: linux-kernel

Hello,
(I hope this is the right place for this - sorry if it is not).

Native Command Queueing (and Tagged Command Queueing) is a feature provided by 
the hardware of newer IDE (and old SCSI) disk drives which basically consists 
in reordering the commands issued on the ATA bus to improve speed.

I assume however that the fastest way to read sectors 101 to 110 is to ask for 
them in that order: 101,102,...,110 . This is a basic assumption made by most 
OSes and apps I presume (otherwise for example DMA performance would be 
catastrophic).

Here is my point: since a bvec consists of _ordered_ requests only, what is 
the use of NCQ ? Requests will arrive to the drive in increasing order, which 
is the best possible ordering performance-wise; thus NCQ will do never do 
anything.

Am I mistaken ? 

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: BIO ordering and NativeCommandQueueing
  2004-06-15 10:02 BIO ordering and NativeCommandQueueing Guillaume Lacôte
@ 2004-06-15 11:36 ` Jens Axboe
  0 siblings, 0 replies; 2+ messages in thread
From: Jens Axboe @ 2004-06-15 11:36 UTC (permalink / raw)
  To: Guillaume Lacôte; +Cc: linux-kernel

On Tue, Jun 15 2004, Guillaume Lacôte wrote:
> Hello,
> (I hope this is the right place for this - sorry if it is not).
> 
> Native Command Queueing (and Tagged Command Queueing) is a feature
> provided by the hardware of newer IDE (and old SCSI) disk drives which
> basically consists in reordering the commands issued on the ATA bus to
> improve speed.
> 
> I assume however that the fastest way to read sectors 101 to 110 is to
> ask for them in that order: 101,102,...,110 . This is a basic
> assumption made by most OSes and apps I presume (otherwise for example
> DMA performance would be catastrophic).
> 
> Here is my point: since a bvec consists of _ordered_ requests only,
> what is the use of NCQ ? Requests will arrive to the drive in
> increasing order, which is the best possible ordering
> performance-wise; thus NCQ will do never do anything.

I think you are confusing scatter-gather with request ordering. And your
terminology is off base - a bvec doesn't consist of ordered requests, it
consist of (max) a single page. A bio consists of bvec's. A request
consits of ordered bio's. The drive queue consist of (fairly well)
ordered requests.

I won't go on about merrits of queueing and depths, search the archives
for that.

-- 
Jens Axboe


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2004-06-15 11:36 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2004-06-15 10:02 BIO ordering and NativeCommandQueueing Guillaume Lacôte
2004-06-15 11:36 ` Jens Axboe

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox