On Mon, Sep 16, 2024 at 05:13:48PM +0000, Farhan Tariq wrote:
> Hello Simon,
> 
> Yes, you’re right. We are indeed targeting hardware offload. Since there
> are limitations when using TCAMs, bypassing them can help optimize both
> lookups and the forwarding process.

Hi Farhan,

First, please don't top-post on OVS mailing lists.

Second, it would be really nice to understand some more information about
the hardware you are working with.

Lastly, and most importantly, regarding the implementation.

Having spoken off-line to others, it is my understanding that the option to
disable megaflows is, at this time, only intended to be used for debugging
purposes.  So if that is a requirement of the design then I think work will
need to be done to make it suitable for non-debugging users.  But perhaps
we can consider some other options.

As it stands a transformation is made from OpenFlow rules to megaflow
datapath rules. But, as you point out, there are use-cases where
exact-match flows make sense. Indeed we see this in the see this in the
user-space datapath. But in that case the approach taken is to implement an
exact-match cache.

I also believe that is it perfectly ok for a datapath implementation to
take accept a megaflow rule and install a finer-grained rule, including an
exact-match rule.  I think this would lead to a similar outcome to your
patch-set, but likely with significantly less modification to OVS.
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to