IETF NVO3 WG Virtual Interim Meeting Agenda 2015-07-10 10:00-11:30 EST Chairs: Benson Schliesser, Matthew Bocci Secretary: Sam Aldrin Minutes: - Agenda 1. WG chairs : Agenda bashing, charter update - Architecture and use case documents are due. Benson will be follow up for completion. - Great deal of work to be done on control plane. - Will be evaluating on how we are doing in terms of milestones, at Prague. 2. NVA Address Mapping Distribution Protocol by Linda Dunbar. - 15min - This draft was discussed in two interims and received comments. - Revised draft was published with suggested sub-TLV’s. - Erik suggested earlier to use OVSDB. - Gave tried description of what OVSDB is. Bhumip> only for config? Linda> Yes - It uses JSON. - Like OVSDB, NVA-NVE defines TLV’s for push model. - Irrespective of OVSDB, the policies are still valid within the draft. - There is no pull model in OVSDB. - Draft defines what triggers to pull and what to pull - If we decide OVSDB in the control plane, the content in the draft is still valid Erik> There is a defined schema as well within OVSDB and used in VXLAN deployment Benson> If correct, there are some other schema which are not present in the RFC Linda> Bitstream is much faster than YANG/JSON Linda> OVSDB could have more things defined, the overhead is quite huge and performance is too slow Linda> Do we need JSON to do control plane or bit stream? Jesse> OVSDB is already adopted and used, so, may not use the newly defined TLV’s. Linda> With OVSDB today, one cannot reach NVA-NVE Jesse> OVSDB does report to controller. Take a look at the published schema. Linda> I have looked at and my judgement is more need to be added to be used a control plane Benson> It would be useful to have mailing list discussion. Info model is needed. Erik> Agree with that. Linda> Consider the info model defined by this draft first. Secondly, consider the schema next. Benson> Good to have mailing list first and take it from there. 3. Virtual Network Transport Protocol by Zhongyu. - 20min - Bhumip is presenting on behalf of Yu. - Draft defines interaction between NVE and NVA - VNTP is based on broad transport mechanisms. - VNTP procedures in various scenarios was explained - Explained various packet formats for VNTP Larry> Why are you running over IP instead of UDP? Bhumip> You could run on UDP. More of operational requirement. Larry> What is the technical reason? Bhumip> No technical reason 4. PMTUD over Vxlan by Vengada Prasad - 10Min - There could be fragmentation in the core and could not convey the mtu size properly - VXLAN G/W may not be able to rely the underlay errors like ICMP errors - David Black reviewed and mentioned that it is good problem to solve. Erik> I have presented a draft on path trace which identifies the MTU value Sam > Why is determining MTU useful and how is it useful when you have hierarchy of transport layers. Erik> It is local decision Sam> Let us take it offline to clarify why this will be useful 5. Path Detection in VXLAN Overlay Network by Dapeng - 15min - Proposed extensions to VXLAN header - If the bit is set the packet response is sent to the controller Sam> Sent comments on the list but no response. Sam> How do you know which controller receives the response? Dapeng> In our deployment, it is one controller connecting to all devices. Sam> Security is a big issue. Please add more content into security section on how this will avoid security related problems like, sniffing out the entire network topology. Dapeng> Will add in the next revision Sam> If a node do not support the new extension, how do you derive it as a unsupported Vs DP failure? Dapeng> Controller has to know that there is unsupported node in the network? Sam> Do add text on how to determine that. Benson> Good discussion. Please find time at Prague and also continue discussion over mailing list. Meeting adjourned. See you all at Prague.