RE: Patch: Asynchronous IPI and e1000 Multiple Queues (aka ReceiveSide Scaling)

Davem brought up the point of how expensive IPIs are.  Our own
performance tests for multi-queue weren't all that impressive...which
just proved his point.  So we decided not to go forward with the patch.
Really, the hardware we coded multi-queue for was not ready for such a
feature.  Hardware that supports MSI-X is really what makes sense for
multi-queue.  Since the 8257x hardware was the first family to support
multi-queue, all previous hardware generations (PCI or PCI-X) would not
have benefited from that patch.

Thanks,
-Jeb

-----Original Message-----
From: Piet Delaney [mailto:piet@bluelane.com]
Sent: Wednesday, August 16, 2006 6:41 PM
To: cramerj; netdev@vger.kernel.org
Cc: Piet Delaney; David Miller; Stephen Hemminger; Subhachandra

Chandra

Subject: Re: Patch: Asynchronous IPI and e1000 Multiple Queues (aka
ReceiveSide Scaling)

On Wed, 2006-08-16 at 18:33 -0700, Piet Delaney wrote:

quoted

I came across your Sept 2005 LKML and NetDev posting:

	http://lwn.net/Articles/152989/

and was wondering what's been going on with this
since your posting. Looks like an interesting
patch that may be useful in our environment.

I just noticed this is for PCI-Express. Has anything
similar been done for PCI-X?

-piet

quoted

-piet

--
Piet Delaney
BlueLane Teck
W: (408) 200-5256; piet@bluelane.com
H: (408) 243-8872; piet@piet.net