Re: [PATCH v2] ipv4: Early TCP socket demux.

[Rick Jones <rick.jones2@hp.com>]

On 06/19/2012 11:14 PM, David Miller wrote:
> From: Eric Dumazet<eric.dumazet@gmail.com>
> Date: Wed, 20 Jun 2012 07:51:29 +0200
>
>> On Wed, 2012-06-20 at 07:49 +0200, Eric Dumazet wrote:
>>
>>> 2) small lived tcp sessions
>>>
>>>     input dst is now dirtied because of the additional
>>> dst_clone()/dst_release()
>>
>> Not realy a concern because we dirty cache line anyway
>>
>> dst_use_noref()
>> {
>> 	dst->__use++;
>> 	dst->lastuse = time;
>> }
>
> Right, the costs probably even out for short TCP flows.
>
> But better to do real tests than to believe what any of
> us say. :-)

netperf -c -C  -t TCP_CC ...  #just connect/close

or

netperf -c -C -t TCP_CRR ...  # with a request/response pair in there

For some definition of "real" anyway :)

rick jones