On Tue, Jan 31, 2012 at 06:00:33PM -0800, Ethan Jackson wrote:
> I'm surprised that the Flow_Table configuration went into the database
> instead of being implemented as OpenFlow extensions. Naively, to me it
> seems like a bit of a layering violation. Can you please briefly
> explain your thinking on this issue?
My thoughts on the division between the two protocols is:
- Anything that doesn't apply to a particular OpenFlow switch
but to the system as a whole goes in the DB. (Doesn't apply
here.)
- Anything that changes potentially very quickly goes in the
OpenFlow protocol. Generally speaking, that is per-flow
state. (Doesn't apply here.)
- Anything specific to a particular OpenFlow connection has to
go in OpenFlow, obviously. (Doesn't apply here.)
- Otherwise, it usually goes in the config protocol.
I'm going to discuss the choice of protocols with Justin, to see what
he thinks, before I push this.
> In ofproto_configure_table() I would think the initial assertion
> should come before we dereference ofproto->tables. Practically
> speaking, it doesn't matter much, just looks a tad strange to me.
Good idea. I made that change.
Thank you for the reviews!
_______________________________________________
dev mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/dev
