Re: [rtcweb] Comments on draft-ietf-rtcweb-stun-consent-freshness-00

["Ram Mohan R (rmohanr)" <[e-mail]>]

Hi Magnus,

Sorry for delay. Please see inline for answers


>-----Original Message-----
>From: Magnus Westerlund [mailto:[e-mail]]
>Sent: Friday, November 15, 2013 2:32 PM
>To: Ram Mohan R (rmohanr);
>[e-mail]; [e-mail]
>Subject: Re: Comments on draft-ietf-rtcweb-stun-consent-freshness-00
>
>Hi,
>
>Removing things where I am in agreement or like your response.
>
>On 2013-11-14 18:23, Ram Mohan R (rmohanr) wrote:
>> Hi Magnus,
>> 
>> Thanks for your comments. Please see inline for responses.
>> 
>> 
>> -----Original Message-----
>> From: Magnus Westerlund <[e-mail]>
>> Date: Tuesday, 12 November 2013 8:28 PM
>
>>>
>>> 3. Section 1:
>>> "This document describes a new STUN usage with a request and response
>>>   which verifies the remote peer consents to receive traffic, and
>>>   detects loss of liveness."
>>>
>>> I am missing a word after "request and response", transaction or
>>> messages maybe?
>> 
>> NEW:
>> This document describes a new STUN usage with a request and response
>> messages which verifies the remote peer consents to receive traffic,
>> and detects loss of liveness.
>
>Isn't it either "which verifies the remote peer's consent to" or "which
>verifies that the remote peer consents to"?


I will change to "which verifies the remote peer's consent to²



>
>
>
>>
>>>
>>> 5. Section 4:
>>>   Consent freshness serves as a circuit breaker (so that if the path or
>>>   remote peer fails the WebRTC browser stops sending all traffic on
>>>   that remote peer), determining session liveness serves the purpose of
>>>   notifying the application of connectivity failure so that the
>>>   application can take appropriate action.
>>>
>> 
>>>
>>> What is the definition of a peer here?
>> 
>> Peer is the sink for the traffic originated from the sender for that
>> flow
>> (5 tuple). If a browser uses different 5 tuple for each stream(Audio,
>> video, data) it sends then it should do consent for each flow. If it
>> uses same 5-tuple (mux case) then there will be only one consent.
>
>Please be explicit about this. I think it is easy to interpret that this
>would apply to all flows from one end-point to a given destination
>address.

Sure. We will add the above text and make it clearer in Section 4


>
>> 
>> 
>> 
>>> What scope does the stop sending
>>> have?
>> 
>> The stream that is being sent on a flow(5 tuple) for which a consent
>> has failed will be stopped. If all the streams are muxed on a same 5
>> tuple the entire traffic will be stopped.
>> 
>
>I am fine with this, as long as the document is explicit about this being
>the scope.

Will make this explicit in the draft


>
>
>>>
>>> 6. Section 4:
>>>   1.  A consent timer, Tc, whose value is determined by the browser.
>>>       This value MUST be 15 seconds.
>>>
>>> I see a contradiction here, should it be determined by the browser or
>>> be
>>> 15 s? If it is 15, can you please motivate why it is this value, or
>>> point to where it is motivated?
>> 
>> The default value of 15 was some thing that EKR and Justin gave us
>> based on the implementation/testing and fine tuning they have done on
>> their browsers. I agree we can have some text around it that explains
>> how this number is arrived. We will add the same.
>
>Good, I think it is important we understand the considerations of why
>this is a reasonable choice.

Sure. Will check with Justin/EKR and add some text that justifies why 15
sec is default.



>
>
>>>
>>> 8. Section 4:
>>>   If a valid STUN Binding Response is not received after 500ms, the
>>>   STUN Binding Request is retransmitted (with the same Transaction ID
>>>   and all other fields).  As long as a valid STUN Binding Response is
>>>   not received, this retransmission is repeated every 500ms until Tf
>>>   seconds have elapsed or a valid response is received.
>>>
>>> So no exponential back-off on the retransmission timer. I think that
>>> is fine. However, I think it do require you to expand a bit on what
>>> happens when one gets multiple response on the same Transaction ID.
>>> For example additional responses do they or do they not reset the Tc
>>>timer?
>> 
>> Additional responses MUST reset the Tc timer.
>> 
>
>Okay, that appears fine.
>
>> 
>>>
>>> Then I wonder about the cases where the RTT is above 500 ms, should
>>> one then scale this factor to be once per RTT?
>
>What about this question?

The ICE agent can learn the RTT of a given path from RTCP reports (if
present). If there is no RTCP reports then it can use STUN BindReq/Res to
learn RTT.

STUN allows you to learn the RTT of a path when you sending Binding
request and get a Binding response for a candidate pair. Before consent
timer
is started the browser¹s ICE agent would have done ICE checks and
concluded on a candidate for each media stream.  So an ICE agent can use
the RTT learnt here.
Thoughts on this ?


>
>>>
>>> 9. Section 4:
>>> "with the same Transaction ID"
>>>
>>> Why not use a new transaction ID for each sent packet? It would allow
>>> tracking loss and determine RTT more reliable. All for the small cost
>>> of keeping track of slightly more Transaction IDs?
>> 
>> Yes, new transaction ID will help to determine the packet loss and RTT.
>> 
>
>Yes, so what is preferable here? Doing cloned retransmission, or
>generating new TIDs for each outgoing request?

Using same Transaction ID would lead the peer ICE agent think this request
as a retransmission and it may not reset its Consent timer.  So I think
using new ID is better.



>
>> 
>> 
>>>
>>> 10. Section 4.
>>>   Liveness timer: If no packets have been received on the local port in
>>>   Tr seconds, the WebRTC browser MUST inform the JavaScript that
>>>   connectivity has been lost.  The JavaScript application will use this
>>>   notification however it desires (e.g., cease transmitting to the
>>>   remote peer, provide a notification to the user, etc.).  Note the
>>>   definition of a received packet is liberal, and includes an SRTP
>>>   packet that fails authentication, a STUN Binding Request with an
>>>   invalid USERNAME or PASSWORD, or any other packet.
>>>
>>> I think this requires some considerations in regards to RTT. If the
>>> RTT is 700 ms, high for a real-time app but not unheard of at all.
>>> Then a Tr = 1 second would fire on a single loss.
>> 
>> Ok.  will start a discussion for this item in a separate email.
>> 

As discussed above the ICE agent can learn the RTT of a given path from
RTCP reports (if present) or compute the RTT from STUN BindReq/Response.
It can then use the learnt value
to influence the Tr value.

Regards,

Ram

>
>Good
>
>Cheers
>
>Magnus Westerlund
>
>----------------------------------------------------------------------
>Multimedia Technologies, Ericsson Research EAB/TVM
>----------------------------------------------------------------------
>Ericsson AB                | Phone  [phone]
>Färögatan 6                | Mobile [phone]
>SE-164 80 Stockholm, Sweden| mailto: [e-mail]
>----------------------------------------------------------------------
>