Re: [rtcweb] Some language on "prioritization"

[Michael Tuexen <[e-mail]>]

On 04 Apr 2014, at 10:51, Harald Alvestrand <[e-mail]> wrote:

> I assume this is the scheduler envisioned in -ndata:
> 
> 
>       SCTP_SS_PRIORITY:  Scheduling with different priorities is used.
>          Streams having a higher priority will be scheduled first and
>          when multiple streams have the same priority, the default
>          scheduling should be used for them.  The priority can be
>          assigned with the sctp_stream_value struct.  The higher the
>          assigned value, the lower the priority, that is the default
>          value 0 is the highest priority and therefore the default
>          scheduling will be used if no priorities have been assigned.
> 
> 
> 
> This sounds like a "strict" scheduler, in that higher priority queues will starve out lower priority ones completely. I remember having the discussion at an IETF meeting about whether we wanted a "strict" scheduler or a "weighted round robin" scheduler for this, but I wouldn't trust my memory with what the conclusion was.
It is.
> 
> Was the conclusion that we should do "strict" scheduling? If so, it may be best to make that consistent across the board - I had written in a "weighted" scheduler for media into the prioritization text that I started this thread with, but I think there's value to consistency.
No, in an IETF WG meeting we discussed this a while ago and the result was a non-strict
priority, weighted fair queueing. This scheduling discipline in not there yet.
> 
> (Note: -ndata has SS_PRIORITY in one place and SS_PRIO / SS_PRIO_INTER in another place. Is there a subtlety here I'm not seeing?)
Yes, there is a difference. It is about interleaving. You can use SS_PRIO
with or without NDATA, SS_PRIO_INTER only with NDATA.
This document still needs some work... Even the constant names might change...

Best regards
Michael
> 
> 
>