On Thu, Dec 03, 2015 at 02:45:15AM +0000, Wenyu Zhang wrote: > > > On 12/2/15, 11:45 AM, "Ben Pfaff" <b...@ovn.org> wrote: > > >On Tue, Dec 01, 2015 at 06:43:49PM -0800, Wenyu Zhang wrote: > >> The patch is to skip BFD packets in ipfix. > >> Bidirectional Forwarding Detection (BFD) packets are for monitoring > >> the tunnel link status and consumed by ovs itself, no need to sample > >> them. > >> Refer to IETF RFC 5881, BFD control packets are the UDP packets with > >> destination port 3784 and BFD echo packets are the UDP packets with dst > >> destination port 3785. > >> Ipfix will skip both BFD control packets and BFD echo packets. > >> > >> Signed-off-by: Wenyu Zhang <wen...@vmware.com> > > > >Does this reduce the accuracy of the estimate of the total number of > >packets? If so, can that be fixed? > > I think that it won¹t affect the accuracy. > Because the total number of packets is per-flow, when the BFD packets are > skipped, > The BFD flow will not exported, but the packets number of other flows > won¹t be affected. > > E.g: the sampling rate is 10%, and there are 100 packets coming, including > 20 BFD packets and 80 packets for other flows. > According sampling action, there may be about 10 packets sampled in > datapath, 2 BFD packets and 8 other packets. > And then in user world ipfix, the 2 BFD packets will be skipped, and we > will estimate of the total number of packets in > Other flow via 8*10=80, it is same as before.
OK. I'm not too happy with the idea of changing what IPFIX bridge-based sampling means, but it does seem like not sampling BFD packets is a reasonable default, and if users want to override that it seems to me that they could use flow-based sampling, which in general gives a lot more control to the user. I added a check that the flow is IP and applied this to master. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
