On Thu, May 26, 2011 at 01:19:18AM -0700, Justin Pettit wrote: > On May 24, 2011, at 4:25 PM, Ben Pfaff wrote: > > > + ovs_be64 packet_count; /* Number of packets in flows. */ > > + ovs_be64 byte_count; /* Number of bytes in flows. */ > > + ovs_be32 flow_count; /* Number of flows. */ > > + uint8_t pad[4]; /* Align to 64 bits. */ > > This isn't directly related to this patch, but it's come up before > that hardware vendors sometimes don't have sufficient counters for > all flows or may not be able to simultaneously gather flow and byte > counters. For port counters, the OpenFlow 1.0 spec explicitly > states that a value of -1 can be used for counters that aren't > supported. I'd suggest that we adopt the same convention for flow > counters. At the very least for our extended ones. What do you > think? It would mostly be the WDP, er ofproto, integrator's job, so > I believe it's only a matter of adding a comment to the description > of the structures.
This sounds reasonable, but integrators never work with these OpenFlow data structures directly, so this isn't the right place to notify integrators. (It is the right place to notify people who are working with our extended data structures, I guess.) The right place would be the comment on rule_get_stats() in ofproto/private.h. When we're aggregating flow statistics, what should be the sum of "unknown" with a known value? I'll make a note to add a patch for this somewhere. _______________________________________________ dev mailing list [email protected] http://openvswitch.org/mailman/listinfo/dev
