[sbm] SMB draft

[Sebastian Moeller <[e-mail]>]

Dear list.

I re-read the current draft and think this elucidates an important issue at the intersection of where application/systems end and were protocols begin.

The draft was a joy to read, I only have two smallish nits:

From the current draft:
Section 2. Introduction
> The document describes the TCP_REPLENISH_TIME socket option for TCP connections using BSD Sockets, and its equivalent for other networking protocols and APIs.¶

Could we maybe explicitly qualify here that this option is a proposal for a potential cross-platform way to create backpressure for local processes? In a sense this is the solution that has been missing so far and also a wake-up call for application developers how to tackle this issue at its root



There is IMHO small "duplication" between

Section 8.2. Packet Expiration
For example, in video conferencing applications it is frequently thought that if a frame is delayed past the point where it becomes too late to display it, then it becomes a waste of network capacity to send that frame at all. However, the fallacy in that argument is that modern video compression algorithms make extensive use of similarity between consecutive frames. A given video frame is not just encoded as a single frame in isolation, but as a collection of visual differences relative to the previous frame. The previous frame may have arrived too late for the time it was supposed to be displayed, but the data contained within it is still needed to decode and display the current frame. If the previous frame was intentionally discarded by the sender, then the subsequent frames are also impacted by that loss, and the cost of repairing the damage is frequently much higher than the cost would have been to simply send the delayed frame. Just because a frame arrives too late to be displayed does not mean that the data within that frame is not important. The data contained with a frame is used not only to display that frame, but also in the construction of subsequent frames.¶

and

Section 8.3. Head of Line Blocking / Traffic Priorities
> For example, in a compressed video stream where a frame is encoded as differences relative to the previous frame, there is no easy way to decode the current frame when the previous frame is missing.¶

These seem quite related, maybe related enough to make this explicit?

Regards
	Sebastian