Re: [MMUSIC] STUN Transactions, still needed in ICE?

[Emil Ivov <[e-mail]>]

On 14.11.14, 6:38, Emil Ivov wrote:
> On 13.11.14, 23:34, Pal Martinsen (palmarti) wrote:
>> Hi,
>>
>> During the TRAM meeting it was briefly discussed if we could changing
>> the STUN transaction ID when retransmitting a STUN request. Main
>> reason is to get more accurate/easier RTT measurements.
>>
>> Can we remove the STUN transaction concept from ICEbis without
>> breaking anything? Would there be simplification there as well as we
>> no longer would need two timers (One transaction timer and one
>> retransmit timer).
>>
>> From a code (At least mine..) perspective it would simplify the STUN
>> part (client/server) and put some more logic into the ICE part.
>
> I guess that for anyone who did implement transactions as per the spec,
> this would be a considerable effort. For us it certainly would be. Much
> more so than the perceived gain of improving RTT measurements.
>
> Wouldn't it be easier to just repurose three or four bits from the
> transaction ID and use them as a counter? This way the receiving side
> can also learn about loss and use it as a factor in constructing pair
> preferences.

And obviously there's also the option of just doing that in an extra 
attribute.

And just to be clear, I do think improving the RTT measurement would be 
a good thing. I just don't believe that repurposing the entire 
transaction ID as a counter is the best way to do it.

Emil


-- 
https://jitsi.org