[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Tue, 07 Jul 2009 13:46:54 -0700
From: Rick Jones <rick.jones2@...com>
To: avorontsov@...mvista.com
CC: David Miller <davem@...emloft.net>,
Andy Fleming <afleming@...escale.com>,
Li Yang <leoli@...escale.com>, netdev@...r.kernel.org,
linuxppc-dev@...abs.org, Lennert Buytenhek <buytenh@...vell.com>
Subject: Re: [PATCH] ucc_geth: Add support for skb recycling
Anton Vorontsov wrote:
> On Tue, Jul 07, 2009 at 11:47:38AM -0700, Rick Jones wrote:
>>Admittedly, all the world is not TCP, but a big chunk is, so are you
>>likely to have reference counts go to zero on the tx queue for
>>anything other than small standalone TCP ACK segments?
>
>
> That's a generic question wrt skb recycling, right? Whether we can
> always recycle transmitted skbs. No, sometimes (or mostly) we can't.
>
> Initially, I was quite puzzled by this support... looking at how
> gianfar driver works (it has the same support as of 0fd56bb5be6455d0),
> I noticed that skb_recycle_check() always returns 0, and so we
> don't recycle the skbs.
>
> Though, things change when the kernel starts packets forwarding,
> *then* skb recycling path actually triggers.
>
> Lennert (skb recycling author) hints us that the gain is indeed
> in forwarding/routing workload:
>
> http://kerneltrap.org/mailarchive/linux-netdev/2008/9/28/3433514
>
>
> Hope I understood everything correctly. :-)
Given the text reads:
This gives a nice increase in the maximum loss-free packet forwarding
rate in routing workloads.
Your understanding is probably correct. Might have been "nice" :) to get a
definition of a "nice increase" though :)
rick jones
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists