I recently returned from IETF 87 and would like to provide a brief synopsis of the event. First, who would have expected the weather to be that hot in Berlin? It hit 96F on Sunday, the first day of the event!
Overall, it was a very well attended event with a very relevant agenda. My first point to make is that the SDN trend continues to gain interest and acceptance. If you’ve been following IETF activity over the last 12 – 18 months or so, you know that this trend started kind of slow but is gaining more traction at each of the IETF events. The future meeting schedule can be found here, and if you’d like a quick review of the IETF 86 event, you can go here.
The NVO3 WG and SDN RG meetings continue to draw the most relevance to the SDN problem space. However, the L3VPN WG and to a lesser extent, the I2RS & PCE WGs, also have correlated activities worth following if SDN is of interest to you.
So, here are the brief updates, organized by WG/RG.
The SDN Research Group kicked off the Monday morning sessions. This RG is chaired by our Brocade SP CTO, Dave Meyer. Dave started the session with a thought provoking presentation that set the bar early. The session included some other quite interesting presentations; such as an SDN-enabled IXP, a presentation on how NFV fits with other IETF SDN activities, and how I2RS and SDN are related (or not).
This was a *very* well attended session, as it was the last time. With the advent of cloud computing and private/public/hybrid data-center clouds, this is where most of that activity is taking place in terms of defining the data-center virtualization problem space, its requirements and eventually the solution space.
The NVO3 architecture team provided an update on their activities. Brocade Principal Engineer, Jon Hudson, is part of this architecture team. This was quite an interactive session. They recommended to promote the current “as-is” VXLAN and NVGRE ID’s to Informational RFC status. This is primarily to document these technologies since they are already implemented and deployed. In addition, other IETF activities relate back to these drafts, so now they are Informational RFCs that can be referenced.
The NVO3 reference architecture that is being developed will identify the key system components and describe how they fit together. This architecture phase will then drive the requirements definition work; and that will then feed into a gap analysis. Some key definitions are being defined; such as, NVE (Network Virtualization Edge) and NVA (Network Virtualization Authority). Some critical ‘on-the-wire’ protocols are being flushed out, such as the NVE to NVA protocol and the Inter-NVA protocol. Finally, consensus on the need for both a push and a pull model of control plane distribution was agreed upon.
The L3VPN sessions are always worth attending, and more recently so since this WG has become closely correlated to the NVO3 activities. Besides the usual discussions of MPLS-related internet drafts and technologies, this session included a very active discussion on how the activities in this WG remains aligned (or not) with the NVO3 activities. This discussion was very much needed, since the last few L3VPN WG sessions included some internet drafts that overlap in some sense with the NVO3 charter. Most of this discussion was around which problem areas should remain in the L3VPN WG and which problem areas should perhaps move to the NVO3 WG. As an example, there was a good amount of consensus that technologies “inside” the DC should belong in the NVO3 WG. However, technologies around “inter-DC” solutions should perhaps remain in this WG. Well, this makes a large assumption that the inter-DC solutions and technologies are based on L3VPN solutions. L3VPN solutions could clearly be one answer to the inter-DC problem, but it’s not the only answer. That’s where it gets fuzzy.
I think one way to clarify this is that if the applicability statement in an ID includes intra-DC problems, then it should be part of the NVO3 WG. If protocol extensions to L3VPN solutions are needed for a particular solution, then clearly that work must be done in the L3VPN WG. But if an ID touches on inter-DC problems, then perhaps it needs to be presented in both WGs; at least until it gets further flushed out. Clear as mud yet?
Another WG that continues to generate a fair amount of activity and interest is PCE. This is also an area of IETF work that is somewhat related to the SDN solution space. This WG is focused on how to enhance traffic-engineering decisions in MPLS networks.
I had two key take-away’s from this session. One is that this WG has quickly moved from working on solutions that only “recommend” traffic-engineered LSPs to the network to also now including solutions that actually “instantiate” those LSPs into the network. In other words, the solutions being discussed here include both the centralized TE control-plane and the actual distributed data-plane. The other important take-away is there is more activity around providing PCE redundancy; that is, how to provide multiple PCE databases for the network and how to keep them synchronized. This is a hard problem to solve and in some sense can help flush out the entire “logically centralized” notion in SDN.
The session ended with an interactive discussion that I thought was quite interesting; well actually, almost amusing. The topic was about whether Auto-BW mechanisms should be pushed back into the network nodes so they can each dynamically adjust their LSP BW. Recall that the primary goal of PCE is to logically centralize the traffic-engineering decisions. It’s all about having a central holistic view of network traffic loads in order to fully optimize the entire network. So, this discussion was about whether it makes sense to then allow each network node to adjust its LSP bandwidth, using auto-BW mechanisms, in a distributed way. Doesn’t this sound counter to the goal of PCE? So, is it a centralized or is it distributed? I hope you see why I thought this entire discussion was somewhat amusing.
I2RS was also very well represented and generated lots of interesting dialogue. I believe this was only the second time this WG met, so the activity here is very early in its definition. The architecture team started off discussing the high-level architecture and a policy framework. All this discussion is centered on the Routing Information Bases (RIBs) of a network node; for example, are multiple RIBs needed? What mechanisms are needed to inject state into the RIB(s)? What mechanisms are needed to extract state from the RIB(s)?
The I2RS activity focuses on the southbound problem space; in other words, the interfaces and protocols needed from the controller or client down to the network node. It does not focus on the northbound interfaces or applications that live on top of the controller.
There was also a question about whether I2RS protocol extensions are being developed in private; or outside the knowledge of this WG. There was a request to encourage people to share those discussions and potential experiments with the general WG to spur discussion.
An interesting I2RS service chaining use case was discussed that is being co-authored by Brocade Principal Architect, Ramki Krishnan.
I dropped into the FORCES session to see how this WG is progressing. This WG has been around for many years but it never really gained much attention in terms of implementation or deployment. Now that SDN is here to stay, my sense is that this WG is trying to re-emerge to become relevant to that conversation.
To that end they have a new charter and some of the terminology being used in this WG is more aligned with the SDN problem space. For example, there are drafts that discuss Virtual Control Element (VCE) nodes and Virtual Forwarding Element (VFE) nodes.
The OPSA WG continued the discussion around a draft presented by Brocade Principal Architect, Ramki Krishnan, on mechanisms for optimal LAG/ECMP link utilization. This capability is important in not only service provider networks but also in research and education networks. Researchers are often the sources for very large IP flow dissemination and the ability to properly load-balance these large flows on LAG bundles is becoming increasingly important. I also presented this draft, on behalf of Ramki, at the recent ESnet Site Coordinators Committee (ESCC) Meeting at the Lawrence Berkeley National Labs and it was very well received by this group.
Regarding the various Routing WGs, here is a short recap.
An interesting presentation in the GROW WG was on a use case of using route reflectors for traffic steering. This idea is not new but it does provide an additional data point on how service providers desire enhanced capabilities to influence traffic patterns in their networks.
Similar to the last IETF, there was more discussion in the IDR WG about the ability to distribute link-state and TE information northbound, using extensions to BGP. This type of capability would allow higher layer applications to make more intelligent traffic-engineering decisions. Kind of sounds SDN-like, doesn’t it?
So, that wraps up this short update on the IETF 87 SP related activities. Please let me know if you have any comments or questions.