Report on presentation of WR at IEEE 802.1
By Rodney Cummings
The following text contains a report on my presentation of White Rabbit
on behalf of Maciej Lipinski at the IEEE 802.1 task group meeting for
AVB Gen 2. The slides and notes for the presentation are now available
on the 802.1
server.
I gave the presentation on the morning of Thursday May 17. I received a
few questions, but I was prepared with answers thanks to Maciej.
Overall, I would say that the presentation was well-received, and
appreciated.
I started the presentation with a brief description of the goals. Although the White Rabbit specs for time distribution are complete, the specs for data distribution are in an early stage. The goals for White Rabbit data distribution align well with the goals for 802.1 AVB Gen 2, which is also in an early stage. Improved communication between the two groups will hopefully enable convergence to a shared standard in the long term.
First, I need to clarify that everything in my report below is my personal interpretation of opinions in the AVB meeting... my "sense of the room" so to speak. Opinions is this sort of working group often evolve as new issues are brought to light. Nothing is truly committed until the amendments to the IEEE 802.1 standard are approved and published (2 to 4 years from now).
That being said, I am optimistic that 802.1 AVB Gen 2 will meet White Rabbit requirements for data distribution. I summarize the topics using Maciej's final Conclusion slide as a guide.
Physical syntonization (time distribution)*
There was interest in the White Rabbit time synchronization protocol. A
few AVB task group members wanted to learn more, and they asked me to
forward a link to the specification. Maciej provided me with a link,
which I have since sent as a message to the IEEE 802.1 listserv.
Multipath solution (topology redundancy)*
WR eLACP: Link Ag is intended to increase bandwidth between two peers.
The new 802.1 standards for Shortest Path Bridging (SPB) are intended to
extend this concept throughout the network, which seems analogous to the
goal of the WR eLACP proposal. The idea of extending SPB for topology
redundancy (as compared to bandwidth) is being discussed in the AVB task
group.
WR eRSTP: During the presentation, members of the audience confirmed that RSTP does indeed have the concept of a backup root, and the WR eRSTP concept of statically configuring backup paths is viable.
The AVB task group has many use cases for topology redundancy, even including applications like home A/V. To help refine the discussion, Andre Fredette of Extreme created a helpful summary.
As with White Rabbit, 802.1 has not agreed on a single solution to the problem, and it is possible that multiple solutions will be provided. The solution shown as 1.a.i. (Setup by protocol, SPB) is somewhat analogous to WR eLACP, and the solution shown as 1.b.i (Traffic engineered w/o protocol, computed by offline tool (human)) is somewhat analogous to WR eRSTP.
Time aware (scheduled) shaper*
I gave a presentation that attempted to move toward decisions for this
project (802.1 Qbv):
here.
The most important decision relates to my Proposal on slide 12 to support multiple windows in bridges as well as talkers. The Cycle and Windows (slots) in each bridge can be distinct. The room agreed that this Proposal is needed in order to optimize bandwidth in large networks (both critical and best-effort traffic).
Static stream / resource configuration*
As we progressed to slides 20-22 of my Scheduled Traffic presentation,
we discussed configuration: static versus dynamic. AVB Gen 1 is
generally dynamic-only, but given the accepted proposal to have distinct
scheduling windows in each bridge, some voiced opinions that AVB Gen 2
should be static-only. My proposal to have a Domain concept in AVB Gen 2
was seen as unnecessary (I agreed), and many felt that AVB Gen 2 did not
require a Stream concept (I agreed). The scheduling windows will be
configured entirely statically, either by flashing bridge defaults, or
by management with SNMP.
This is great news for White Rabbit, because it means that AVB Gen 2 will support completely static configuration.
It does raise some challenges for some industrial applications, which historically expect a certain degree of dynamic configuration. I agreed to create a presentation on this for the next AVB meeting in July, including a proposal for dynamic configuration of the windows (without a stream concept). Nevertheless, even if I succeed in getting dynamic configuration supported as an option, it is clear that static configuration will be supported as well.
VLAN-wide broadcast*
The direction of the AVB task group is less clear on this topic. When I
presented this on slides 21-22 of my Scheduled Traffic presentation,
some voiced opinions that it would be fine to support both broadcast
(VLAN filter) and multicast (MAC filter). I'd say that the room
generally disagreed with the idea that VLAN filtering is faster for
failover (they handle both quickly). The rationale for allowing
broadcast was more "sure... why not?". Nevertheless, when I presented
the question again during my presentation of White Rabbit, there seemed
to be some opposition to broadcast. I'll continue to promote broadcast,
and we will see how it goes in future meetings.
Preemption*
We did not discuss this topic much due to open issues in coordination
with IEEE 802.3. We intend to sync up with 802.3 during the plenary
meeting in July.
Thanks for Maciej for his efforts in creating the slides. I'm very excited about the prospects as AVB Gen 2 evolves.