Re: [PATCH net-next 3/3] net: allow to leave the buffer fragmented in skb_cow_data()

[Mathias Krause <mathias.krause@secunet.com>]

On 07.11.2013 10:01, Herbert Xu wrote:
> On Thu, Nov 07, 2013 at 09:55:34AM +0100, Mathias Krause wrote:
>> Indeed. I want to avoid the costly memcpy() on the CPU serving the NIC
>> interrupt, as that is a bottleneck in my setup. The packet processing --
>> encrypting/decrypting of ESP packets -- gets mostly parallelized via
>> pcrypt, so that's fine. But the initial network processing, i.e. getting
>> to pcrypt, is what's throttling the throughput currently. (RPS only
>> partly solves this problem as for the ESP receive path most traffic ends
>> up on the same flow).
> 
> How about this: instead of doing in-place encryption/decryption
> let's allocate a new destination buffer and encrypt/decrypt into
> it.  Then we essentially get the copy for free.
> 
> So instead of modifying skb_cow_data you can change esp4.c, etc.

That might be another option. But I fear it might affect performance due
to caching or NUMA effects. :/

I'll think about it and will try to investigate both approaches.


Thanks,
Mathias