- From: Harald Alvestrand <harald@alvestrand.no>
- Date: Wed, 13 Jun 2012 09:30:03 +0200
- To: Martin Thomson <martin.thomson@gmail.com>
- CC: public-webrtc@w3.org
On 06/13/2012 09:21 AM, Martin Thomson wrote:
> On 13 June 2012 00:07, Harald Alvestrand<harald@alvestrand.no> wrote:
>> How far down do you think we have to drive the setup time before you would
>> not call it "abysmal"?
> Responsiveness at call setup and responsiveness when disabling hold
> probably have different user expectations. With the number of round
> trips required, I can imagine having to restart ICE and handshake
> taking a fairly long time. Especially with a longer end-to-end RTT.
Sure - once I have a target number, I can divide the required number of
round trips by that and get the maximum RTT we can support without a
redesign - that's the way to know if I'm in trouble or not.
If we agree that 2s is an unproblematic call setup time, 0.5s is an
unproblematic call reestablishment time on interface changeover, and 1s
is an unproblematic reestablishment time after call idle (just pulling
some numbers out of thin air), we can take those figures and see at what
RTTs we get into trouble for our various designs.
>
> Is your 2s statistic based on an intercontinental RTT?
This was a measurement, and one measurement is not a statistic :-)
It was based on a ~20 ms RTT to the signalling server
(apprtc.appspot.com) and a very local connection between the endpoints
(2 tabs in the same browser). Close to the best possible case!
Harald
Received on Wednesday, 13 June 2012 07:30:34 UTC