To answer my own question: I wasn't specifying the in_port when calling of.ofp_match.from_packet, so in_port was wildcarded. Wildcard matches go into the linear search table instead of the hash search table - makes sense.
Specifying the in_port fixed my problem. I'm loving this OpenFlow stuff! -Russ On Wed, Oct 10, 2012 at 11:18 AM, Russ Weeks <rwe...@newbrightidea.com>wrote: > By the way, here's the output of dpctl dump-tables: > > root@OpenWrt:/# dpctl dump-tables tcp:127.0.0.1:6634 > stats_reply (xid=0xe724908a): flags=none type=3(table) > 2 tables > 0: hash2 : wild=0x00000, max= 32768, active=0 > lookup=126035, matched=0 > 1: linear : wild=0x3fffff, max= 100, active=96 > lookup=126035, matched=100564 > > Apparently none of my flow rules are going into the hash2 table. Maybe > this is due to the match criteria that POX is specifying? I'm just using > the forwarding.l2_learning module right now. > > -Russ > > On Wed, Oct 10, 2012 at 11:13 AM, Russ Weeks <rwe...@newbrightidea.com>wrote: > >> Hi, All, >> >> I've deployed OpenFlow 1.0 on a Mikrotik RouterBoard 450g based on the >> excellent online instructions[1]. I've hooked it up to the POX controller >> and I can see that the flows are being provisioned, and everything is >> working great. But as I stress the network a little bit with numerous HTTP >> connections I frequently see the error "OFPFMC_ALL_TABLES_FULL" in response >> to a flow mod message. Running dpctl dump-flows on my switch shows roughly >> 100 active flows. >> >> Since this is a user-space software switch, am I right in thinking that >> these flow rules are kept in the ofdatapath process heap? (ie. this is not >> a hardware limitation) And since I've got 256MB of RAM on this device, is >> there maybe a CLI parameter I can use to increase the table size? >> >> Thanks very much, >> >> -Russ >> >> 1: http://www.openflow.org/wk/index.php/Pantou_:_OpenFlow_1.0_for_OpenWRT >> > >
_______________________________________________ openflow-discuss mailing list openflow-discuss@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
