[ovs-discuss] About the tuple space search

Justin Pettit jpettit at nicira.com
Wed Dec 3 02:13:38 UTC 2014


On Dec 2, 2014, at 5:07 PM, HyunYong Lee <ifjesus7 at gmail.com> wrote:
> 
> I am trying to understand Open vSwitch, but I am not sure about something.
> So, I would like to ask some questions.
> 
> First, do megaflows support the exact match?

They're as precise as they need to be to enforce the policy, but it's unlikely a packet will need to match on every field in a flow.  In the Linux kernel, there is a per-CPU exact match cache that sits in front of the megaflow tables that can make lookup faster.

> If not, how does the search in the cache happen?

Assuming it misses in the cache, we look in each of the tuple tables.  Since we guarantee in userspace that they will not be overlapping, we can just use the first match that we find.

> Second, if we can know a proper key for a given packet in the tuple space search with high probability, how much performance improvement can we expect?

I'm not sure that I understand the question.  How would you know that before you do the search?  One thing that we haven't done that might be useful is to reorder the tuples based on frequency of being hit.  I don't believe that's been implemented yet, but the exact-match cache helps for commonly seen packets.

--Justin





More information about the discuss mailing list