call 3/14/2015 draft notes

[Jouni Korhonen <jouni.korhonen@xxxxxxxxxxxx>]

Folks,

Start:	3/17/2015 6AM PDT
End:	3/17/2015 7AM PDT

Present:
	Jouni Korhonen
	Richard Tse
	Peter Ashwood-Smith
	Igor Freire
	Bomin Li
	Nader Zain
	Gareth Edwards
	Patrick Diamond
	Steinar Bjornstad
	Kevin Bross
	Duncan Cockburn
	Zakarin Tayq
	Satory Matsushima

Agenda:
* Next upcoming FtF meeting
* Strawman proposal for the RoE header

Discussion:
* Normal reminders about the coming meeting and the typical document deadline (10 days before the meeting start).
* The goal would could up with the initial structure of the document and even some content in there.


* We went through the strawman proposal (by Jouni) for the RoE header (attached).
* Most of the header is obvious.
* A comment from Peter (relates to the IEEE 802.1 meeting last week) that in CPRI the jitter is not in the context of clock recovery but PDV. Jouni/Peter followed by a short summary from the IEEE 802.1 meeting (about the transport profile proposal and CPRI cooperation presentation).
* Jouni explained that each type of the flow (control packets, antenna carriers, vendor specific data, structure agnostic containers, ..) will all reserve one packet type. 32 types may feel plenty but it is still a scarce resource.
* flow_id is for multiplexing purposes on the endpoints. Initial 128 looks many but considering upcoming radio enhancements (e.g. Massive MIMO) it does not sound too large.
* Comment from Peter that transport layer means for multiplexing could also be used (VLANs, burn more MAC addresses, MPLS labels, etc). Jouni preferred more transport independent way of doing the same. Something to agree on later.
* Unused bits specifically between packet type and flow ids to allow growing both if needed.
* Most of the time was spent on the timestamp discussion. At the moment (in the strawman proposal) that is a presentation time and to be determined whether that is ok.
* Discussion on the frequency of timestamping. The use of "t-bit" could be replaced with packet types (and giving one more bit for packet types).
* It was noted that RoE endpoints are assumed to be already time/freq/phase synchronized by other means. The timestamp (and sequence numbers) here is to help reconstruct the packet streams & play them out from jitter buffer rather than helping with the functions e.g. 1588 already does.
* Need to decide whether 1ns accuracy is actually enough. This needs to be verified. So far 1ns looks enough. Patrick comments about the challenges of sub 10ns accuracy with 1588 (and other technologies) and points out the ongoing high accuracy work in 1588 (that goes into picosecond level).
* AP to Patrick to distribute pointers to high accuracy work.
* Eventually the standard has to spell out exactly how the timestamp is to be used, how that relates to ref time and used overall sync solution.

* Running out of time.. discussion to move to the reflector.

- Jouni

Attachment: header.docx
Description: header.docx