[ovs-dev] [PATCH v2] datapath: Remove tunnel header caching.

[Jesse Gross]

On Fri, Oct 26, 2012 at 11:21 AM, Ben Pfaff <blp at nicira.com> wrote:
> On Fri, Oct 26, 2012 at 11:19:37AM -0700, Ben Pfaff wrote:
>> On Thu, Oct 25, 2012 at 11:14:38AM -0700, Pravin B Shelar wrote:
>> > v1-v2:
>> >     - Fixed capwap fragment case.
>> >     - simplified tnl_send.
>> > --8<--------------------------cut here-------------------------->8--
>> >
>> > Tunnel caching was added to reduce CPU utilization on TX path
>> > by caching packet header, So performance gain is directly proportional
>> > to number of skbs transferred. But with help of offloads skb are getting
>> > larger. So there are less number of skbs. Therefore header caching does
>> > not shows similar gains we seen in past. And now kernel 3.6 has removed
>> > dst caching from networking which makes header caching even more tricky.
>> > So this commit removes header caching from OVS tunnelling.
>> >
>> > Signed-off-by: Pravin B Shelar <pshelar at nicira.com>
>>
>> Is there any benefit in adding a comment to tunnel.h remarking that
>> the bit 1<<6 is reserved because it was previously used?  (Maybe not,
>> if we're deleting this interface entirely in another version.)
>
> Oh, and in netdev-vport.c, should we silently ignore the header_cache
> option instead of warning that it's unknown?  Or instead warn that
> it's not useful anymore?

We warn for all unknown options so it should happen automatically.
It's also probably not that important for header caching because it's
really supposed to be a transparent optimization.  Not having it
shouldn't change any behavior.