Re: [PATCH 7/8] [PATCH] dynamic calculation of event message size for ctnetlink

[Patrick McHardy <kaber@trash.net>]

Pablo Neira Ayuso wrote:
> Patrick McHardy wrote:
>> Pablo Neira Ayuso wrote:
>>> Pablo Neira Ayuso wrote:
>>>> Patrick McHardy wrote:
>>>>> These calculations look somewhat expensive to perform for every
>>>>> message.
>>>>> Do you have any numbers for this new patch that shows the difference
>>>>> in CPU usage compared to the resizing done by af_netlink.c?
>>>> Fabian Hugelshofer reported some reduction (~5%) on an embedded
>>>> environment but he was using top to measure the difference. I'll collect
>>>> some more trustable data and get back to you.
>>> Some oprofile results:
>>>
>>> wo/patch
>>> 2189      0.0305  nf_conntrack_netlink.ko  nf_conntrack_netlink
>>> ctnetlink_conntrack_event
>>>
>>> w/patch
>>> 2302      0.0440  nf_conntrack_netlink.ko  nf_conntrack_netlink
>>> ctnetlink_conntrack_event
>>>
>>> While __alloc_skb and netlink_broadcast report similar values for w/ and
>>> wo/ the patch.
>> So its actually getting worse? :) Any other differences, like less
>> cycles for memcpy in netlink_trim()?
> 
> netlink_trim is inlined, so it is included in netlink_broadcast, and
> there's no improve in memcpy nor netlink_broadcast. I'm going to repeat
> all the test to check if I'm doing something wrong, until that, let's
> keep it back.

Thats really strange, there has to be at least some reduction of work
because we should be avoiding the packet copy in netlink_trim. Unless
there's a bug somewhere in the calculation and we're still
overallocating by more than 50%.