Re: [rtcweb] Non-media data service consensus and requirements

[Bernard Aboba <[e-mail]>]

I do not support an unreliable datagram service that can be used to send arbitrary data.  

For example, it seems dangerous for a Web browser under the control of an attacker to be able to send RIP, SNMP or DNS packets to arbitrary destinations. 

For these transactional exchanges the overhead of ICE would be excessive and so there will be a very strong temptation to cut corners. 

Assuming that the goal is not to send arbitrary data, then we need to dig into the transport requirements more. 

For example, is the non-media data to be synchronized with media (e.g. real-time text)?

Is there a session associated with the non-media data (e.g. XMPP or MSRP exchanges)? 

Is there a reliability requirement? 

Is it congestion-controlled? 

How long-lived are the flows? 




---------------------------------------------------
From: [e-mail]
To: [e-mail]
Date: Mon, 27 Jun 2011 09:36:30 +0200
Subject: [rtcweb] Non-media data service consensus and requirements

WG,
 
At the interim it was planned to have a bit discussion on the datagram
service for RTCWEB. The first question to try to resolve if there
is consensus for including some form of non real-time media (i.e. not
audio, video) service between peers. This is a bit tangled with the
actual requirements and use cases. But there was views both for it and
against it on the mailing list. So lets continue and try to come to a
conclusion on this discussion.
 
The use cases mentioned on the mailing list are:
 
- Dynamic meta data for Conference and other real-time services
 
- Gaming data with low latency requirements
 
Does anyone like to add additional use cases?
 
Based on my personal understanding this points to primarily have the
RTCWEB provide a unreliable datagram service. This clearly needs
additional requirements to be secure and safe to deploy, but more about
this below. I still like to ask the WG here a question.
 
Are you supporting the inclusion of a unreliable datagram service
directly between peers? Please provide your view and any additional
statements of motivation that you desire to provide.
 
Secondly, there is a question if there needs to have something that
provides reliable message (of arbitrary size) or byte stream oriented
data transport between the peers. I personally foresee that people will
build JS libraries for this on top of a unreliable datagram service. If
you desire reliable data service as part of the standardized solution
please provide motivation and use case and requirements.
 
I also want to take a stab on what I personally see as the requirements
that exist on unreliable datagram service in the context of RTCWEB.
 
- Unreliable data transmission
- Datagram oriented
   * Size limited by MTU
     - Path MTU discovery needed
   * Fragmentation by the application
- Low latency, i.e. Peer to Peer preferable
- Congestion Controlled, to be
   * Network friendly
   * Not become a Denial of Service tool
- Security
  * Confidentiality
  * Integrity Protected
  * Source Authenticated (at least bound to the signalling peer)
  * Ensure consent to receive data
 
Please debate the above. This is an attempt to ensure that we can
establish WG consensus on both data service and any requirements.
 
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]
----------------------------------------------------------------------