Sounds right. -- Murphy
On Oct 22, 2013, at 2:12 PM, durga <c.vijaya.du...@gmail.com> wrote: > Thanks Murphy! > So can it be said - the difference is in how the flow table is affected. > ofp_flow_mod() alters the flowtable entries, where as ofp_packet_out() > message instructs the switch to do - rather than installing/modifying the > flows in the flow table. If yes, it does clear up few things! > > > > Cheers! > Durga > > > > On Wed, Oct 23, 2013 at 7:38 AM, Murphy McCauley <murphy.mccau...@gmail.com> > wrote: > On Oct 22, 2013, at 1:23 PM, durga <c.vijaya.du...@gmail.com> wrote: > >> Can anyone help me understand in what scenarios which of the openflow >> messages have to be used? I understand ofp_flow_mod() is to modify flow >> table entries on the switch ( I have been using this message all along) and >> ofp_packet_out()(which I don't know why to use) is when the controller wants >> the switch to send a packet, which can also be achieved by modifying the >> flow table entries using ofp_flow_mod() message. Are there any other factors >> which decide which message to be sent when? > > ofp_flow_mod is meant to install table entries. It has sort of a special > shortcut which allows sending a packet which has been buffered at the switch, > since this was expected to be a common case (for reactive controllers which > install table entries in response to packet_ins). > > ofp_packet_out is meant to cause a switch to send a packet. This may be one > which has been buffered at a switch or one that hasn't been. It may have > originally come from a switch, or it may be generated by the controller. For > example, the LLDP packets sent by discovery couldn't be sent via an > ofp_flow_mod, since they were never on the wire. Even when a packet comes > via a packet_in, you may want to resend it without installing a table entry > for it, which would make ofp_flow_mod unsuitable. l2_pairs does this, for > example, when it doesn't know the destination -- it uses a packet_out to > flood the packet. It doesn't want to do this permanently, so it doesn't > install a table entry that does it -- it just does it the once. > > (POX confuses matters slightly because libopenflow's ofp_flow_mod may > actually generate an ofp_packet_out too in order to simplify a particular > case, but this can largely be ignored.) > >> Also,a note to admin.is there a way I can scna through previous queries >> posted by other members? > > This page has a link to the mailing list archive: > http://www.noxrepo.org/community/mailing-lists/ > >