[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20150102220107.GA28599@gondor.apana.org.au>
Date: Sat, 3 Jan 2015 09:01:07 +1100
From: Herbert Xu <herbert@...dor.apana.org.au>
To: David Miller <davem@...emloft.net>
Cc: eric.dumazet@...il.com, thomas.jarosch@...ra2net.com,
netdev@...r.kernel.org, edumazet@...gle.com,
steffen.klassert@...unet.com, bhutchings@...arflare.com
Subject: Re: tcp: Do not apply TSO segment limit to non-TSO packets
On Fri, Jan 02, 2015 at 03:36:55PM -0500, David Miller wrote:
> From: Eric Dumazet <eric.dumazet@...il.com>
> Date: Fri, 02 Jan 2015 10:24:00 -0800
>
> > Non TSO/GSO path is known to be better for devices unable to perform TX
> > checksumming, as we compute the checksum at the time we copy data from
> > user to kernel (csum_and_copy_from_user() from tcp_sendmsg())).
>
> Non-SG capable devices suffer in this scenerio as well.
Yes I was referring to using GSO on non-SG/non-checksumming devices.
Anything that supports checksum/SG should obviously be using GSO.
IIRC when I first tested this GSO is basically on par for the non-SG
case as the overhead of the extra copying was offset by the benefit
of a larger MTU through the stack.
So has anyone actually observed worse performance with GSO on these
devices (you could even test on a GSO-capable device simply by
disabling checksumming)?
Also the fact that this bug took two years to surface means that
very few people are actually using non-GSO in the real world as
this bug is easily triggered by a PMTU event.
Cheers,
--
Email: Herbert Xu <herbert@...dor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
--
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