Re: [tsvwg] thoughts on operational queue settings (Re: CC/bleaching thoughts for draft-ietf-tsvwg-le-phb-04)

[Toerless Eckert <[e-mail]>]

On Thu, Apr 12, 2018 at 10:14:00AM +0200, Mikael Abrahamsson wrote:
> On Thu, 12 Apr 2018, Toerless Eckert wrote:
> 
> > I would hope today we could get a separate WFQ for LE on all
> > platforms which looks to me like the much easier and easier to
> > predict solution. If you tell me thats not the case, my
> > interest to delve again into RED profiles will raise again *sigh*
> 
> There are too many platforms and they're all different as you already said.
> Common platforms will give you the possibility of having a few queues, so I
> will also investigate what the result would be of having three queues, one
> for LE, one for BE and one for rest, where LE gets 5% and the two other
> queues share the rest (getting half the bw each if both are congested).
> 
> This kind of setup would result in no congestion seen by rest of traffic
> even if BE and LE are fully congested.

Right. Whats "the rest" ?

Stuff like IPTV/voice/guaranteed L3VPN service traffic or the like ?

> If I set the LE drop profile to 1ms/2ms then it stopped working when there
> was BE congestion. I think this is not what one wants to happenn for normal
> customers, instead LE traffic should get some bandwidth even if it's a
> smallish single digit percentage of the total.

Argument in favour of some small > 0 share of bandwidth:

It does allows for many apps to continue in LE working without change
because they would just become just terribly slow - but would not need to
be restarted/rewritten which is necessary if the rate would go to
0 for periods longer than common TCP timeouts.

The other interesting question for the LE queue is what impact
the queue size has. If i remember correctly, the queue size
is not really in msec even if it can be specified in msec, right ?

Instead it would be a translation of msec to an actual #packet/#byte amount
for the nominal share of the queue, right ? (i am not quite sure, would
need to check docs).

I am asking, because if that interpretation is right, and queue size
of LE is e.g.: 20 msec at 5% share of the link, but LE actually
get 100% of link when BE is unused, then queue size in that sitution
would be just 1 msec, and that might have a negative impact.

Cheers
    Toerless