Re: [ledbat] other open issues

On Fri, Jun 11, 2010 at 11:25 AM, Padma Bhooma <bhooma@apple.com> wrote:
> 1. How will base history expire after some idle time? May there should be a recommendation to throw away the history and re-learn delay on the link.

Padma,

Good point, added.

> 2. The linear controller that LEDBAT uses suggests that the congestion window should be increased proportional to the change in one-way delay. This is very good for responding quickly to changes in delay. But this can cause the window to increase at a faster rate than TCP would do with its probing mechanism using additive-increase. Although the maximum increase in window is capped to be less than Best-effort TCP, the rate of increase can be more. Does this concern anyone?

The change is not proportional to the change in one-way delay.

It is proportional to the difference between (non-negative) queuing
delay estimate and the target:

"The maximum speed with which we can increase our congestion window is
then when queuing delay estimate is zero. To be on the safe side,
we'll make this speed equal to how fast TCP increases its sending
speed.  Since queuing delay estimate is always non-negative, this will
ensure never ramping up faster than TCP would."

> Some of these decisions can be made implementation-specific. I think, it will be good if there is some discussion about it.
> Thanks,
> Padma
>
> On Jun 8, 2010, at 12:00 PM, ledbat-request@ietf.org wrote:
>
> Date: Tue, 8 Jun 2010 16:03:37 +0200
> From: "Rolf Winter" <Rolf.Winter@neclab.eu>
> Subject: [ledbat] Open issues, part 1
> To: <ledbat@ietf.org>
> Message-ID: <547F018265F92642B577B986577D671C014B9A15@VENUS.office>
> Content-Type: text/plain; charset="us-ascii"
>
> Hi,
>
> Here is the first Email in a series of Emails (to keep Email threats separate) which deal with open points in the congestion control draft. I pulled those together from various sources such as meeting minutes. I would like have a documented agreement on them. If there are other open question please speak up.
>
> 1. Parameters
> There were comments on the parameterization of the congestion control draft. The main concern was that they seem to come out of thin air. To close this issue I'd like to ask the group whether they think the parameters are sensible or not (e.g. 25ms delay target). It would be good to hear from people such as, say, game developers what they have to say about this target. A better description of why these were chosen is surely something that should go into the text.
>
> Best,
>
> Rolf
>
> NEC Europe Limited | Registered Office: NEC House, 1 Victoria Road, London W3 6BL | Registered in England 2832014
>
>
>
> _______________________________________________
> ledbat mailing list
> ledbat@ietf.org
> https://www.ietf.org/mailman/listinfo/ledbat
>

--
Stanislav Shalunov
BitTorrent Inc
shalunov@bittorrent.com

personal: http://shlang.com