As David alluded, this is probably worth noting as an informational guideline - this is perhaps not a data plane requirement.
This problem is not unique to OVS or Overlays. Besides route flapping, out-of-order packet delivery can happen during mac learning in the current switches. There are many possible solutions, each one of these have their tradeoffs. Thanks, Ramki From: [email protected] [mailto:[email protected]] On Behalf Of Lizhong Jin Sent: Sunday, May 05, 2013 7:46 PM To: Black, David Cc: [email protected] Subject: Re: [nvo3] The packet sequence problem in NVO3 Thanks David. Since the draft-ietf-nvo3-dataplane-requirements<http://tools.ietf.org/wg/nvo3/draft-ietf-nvo3-dataplane-requirements/> is already a WG draft, we need the WG consensus to add this new requirement. Any solutions deployed with openvswitch should care about the potential influence of packet out-of-order. Regards Lizhong On Mon, May 6, 2013 at 3:42 AM, Black, David <[email protected]<mailto:[email protected]>> wrote: My initial reaction is that this seems no worse than the typical route-flapping effects of some routing updates - it's certainly worth noting, but I would not require that it be eliminated completely in all solutions. Thanks, --David ---------------------------------------------------- David L. Black, Distinguished Engineer EMC Corporation, 176 South St., Hopkinton, MA 01748 +1 (508) 293-7953<tel:%2B1%20%28508%29%20293-7953> FAX: +1 (508) 293-7786<tel:%2B1%20%28508%29%20293-7786> [email protected]<mailto:[email protected]> Mobile: +1 (978) 394-7754<tel:%2B1%20%28978%29%20394-7754> ---------------------------------------------------- From: [email protected]<mailto:[email protected]> [mailto:[email protected]<mailto:[email protected]>] On Behalf Of Lizhong Jin Sent: Friday, May 03, 2013 9:51 PM To: [email protected]<mailto:[email protected]> Subject: [nvo3] The packet sequence problem in NVO3 Hi all, Recently, I found the packet sequence problem in openVswitch(with VXLAN), please refer to the detail http://openvswitch.org/pipermail/discuss/2012-April/007002.html. Although this is caused by the specific openvswitch implementation, but such kind of implementation does gain better performance, while the side effect is the sequence problem. Then I suggest to consider the sequence requirement in NVO3-dataplane, and try to provide some solutions in the later stage. Regards Lizhong
_______________________________________________ nvo3 mailing list [email protected] https://www.ietf.org/mailman/listinfo/nvo3
