From mboxrd@z Thu Jan 1 00:00:00 1970 From: Rick Jones Subject: Re: [PATCH] Super TSO v3 Date: Tue, 24 May 2005 09:12:21 -0700 Message-ID: <42935265.3040103@hp.com> References: <20050524003208.GA25778@gondor.apana.org.au> <20050523.192917.48530622.davem@davemloft.net> <20050524023256.GA29242@gondor.apana.org.au> <20050523.193817.112290763.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: To: netdev@oss.sgi.com In-Reply-To: <20050523.193817.112290763.davem@davemloft.net> Sender: netdev-bounce@oss.sgi.com Errors-to: netdev-bounce@oss.sgi.com List-Id: netdev.vger.kernel.org David S. Miller wrote: > From: Herbert Xu > Date: Tue, 24 May 2005 12:32:57 +1000 > > >>True, the Nagle algorithm itself aims to do something different >>from this function. However, the act of turning Nagle off is >>an indication that the application wants to minimise the latency >>by sending things out ASAP. So we should either respect that >>here by not delaying the packets to increase the TSO size, or >>we'll need a new socket option to do that for TSO. > > > Sure, we can check tp->nonagle to turn this deferring off. > > But, I bet there are folks who want traditional Nagle turned > off, yet TSO chunking enabled. I'm not sure there will be very many of those. I would have thought the folks who turn-off Nagle are typically doing small sends. If they were not wanting to wait to aggregate their small sends into MSS segments, or wait for the RTT, I doubt thet are looking to wait to aggregate to something larger than the MSS. rick jones