Internet DRAFT - draft-jennings-dispatch-snowflake
draft-jennings-dispatch-snowflake
Network Working Group C. Jennings
Internet-Draft S. Nandakumar
Intended status: Standards Track Cisco
Expires: September 3, 2018 March 2, 2018
Snowflake - A Lighweight, Asymmetric, Flexible, Receiver Driven
Connectivity Establishment
draft-jennings-dispatch-snowflake-01
Abstract
Interactive Connectivity Establishment (ICE) (RFC5245) defines
protocol machinery for two peers to discover each other and establish
connectivity in order to send and receive Media Streams.
This draft raises some issues inherent in the assumptions with ICE
and proposes a lightweight receiver driven protocol for asymmetric
connectivity establishment.
Status of This Memo
This Internet-Draft is submitted in full conformance with the
provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering
Task Force (IETF). Note that other groups may also distribute
working documents as Internet-Drafts. The list of current Internet-
Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months
and may be updated, replaced, or obsoleted by other documents at any
time. It is inappropriate to use Internet-Drafts as reference
material or to cite them other than as "work in progress."
This Internet-Draft will expire on September 3, 2018.
Copyright Notice
Copyright (c) 2018 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
(http://trustee.ietf.org/license-info) in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with respect
to this document. Code Components extracted from this document must
Jennings & Nandakumar Expires September 3, 2018 [Page 1]
Internet-Draft snowflake March 2018
include Simplified BSD License text as described in Section 4.e of
the Trust Legal Provisions and are provided without warranty as
described in the Simplified BSD License.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2
2. Terminology . . . . . . . . . . . . . . . . . . . . . . . . . 3
3. Problem Statement . . . . . . . . . . . . . . . . . . . . . . 3
4. Snowflake for connectivity establishment . . . . . . . . . . 4
4.1. System Components . . . . . . . . . . . . . . . . . . . . 4
4.2. Protocol Workings . . . . . . . . . . . . . . . . . . . . 4
4.3. Advantages of Snowflake . . . . . . . . . . . . . . . . . 7
4.3.1. Diagnostics . . . . . . . . . . . . . . . . . . . . . 7
4.3.2. Timing . . . . . . . . . . . . . . . . . . . . . . . 7
4.3.3. Asymmetric Media . . . . . . . . . . . . . . . . . . 8
4.3.4. Fast Start . . . . . . . . . . . . . . . . . . . . . 8
5. IANA Consideration . . . . . . . . . . . . . . . . . . . . . 8
6. Security Considerations . . . . . . . . . . . . . . . . . . . 8
7. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 8
8. References . . . . . . . . . . . . . . . . . . . . . . . . . 8
8.1. Normative References . . . . . . . . . . . . . . . . . . 8
8.2. Informative References . . . . . . . . . . . . . . . . . 8
Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 9
1. Introduction
ICE was designed over a decade ago and certain assumptions about the
network topology, timing considerations, application complexity have
drastically changed since then. Newer additions/clarifications to
ICE in [I-D.ietf-ice-rfc5245bis] and Trickle ICE
[I-D.ietf-ice-trickle] have indeed help improve its performance and
the way the connectivity checks are performed.
However, enforcing stringent global pacing requirements coupled with
brute force connectivity checks, tightly coupled timing dependencies
between the ICE agents, the need for symmetric connection setup, for
example, has rendered the protocol inflexible for innovation and
increasingly difficult to apply and debug in a dynamic network and
evolving application contexts.
This specification defines Snowflake, where, like ICE, both sides
gather a set of address candidates that may work for communication.
However, instead of both sides trying to synchronize connectivity
checks in time-coupled fashion, the sending side acts as a slave and
sends STUN packets wherever the receiving side tells it to and when
it is told to do so. The receiving side is free to choose whatever
algorithm and timing it wants to find a path that works. The sender
Jennings & Nandakumar Expires September 3, 2018 [Page 2]
Internet-Draft snowflake March 2018
and receiver roles are reversed for media flow in the opposite
direction.
The current version of this draft builds on its original
instantiation submitted in year 2015 as
<https://datatracker.ietf.org/doc/draft-jennings-mmusic-ice-fix/>
2. Terminology
In this document, the key words "MUST", "MUST NOT", "SHOULD", "SHOULD
NOT", "MAY", and "OPTIONAL" are to be interpreted as described in RFC
2119 [RFC2119] and indicate requirement levels for compliant
implementations.
3. Problem Statement
ICE was developed roughly ten years ago and several things have been
learned that could be improved:
1. It is spectacularly difficult to debug and analyze failures or
successes in ICE or develop good automated tests. Many
implementations have had significant bugs for long periods of
time. This is further complicated by the timing dependency as
explained next.
2. It is timing dependent. It relies on both sides to to do
something (candidate pairing, validation) at roughly the same
time and that ability to do this goes down with the number of
interfaces and candidates being handled. Mobile interfaces, dual
stack agents make this situation worse.
3. Differences in interpretation and implementation of the protocol
with respect to aggressive vs normal nomination may hinder rapid
convergence or may end up in agents choosing suboptimal routes.
4. It does not discover asymmetric routes. For example UDP leaving
a device may work just fine even though UDP coming into that
device may not work at all.
5. Many deployments consider using a TURN/Media Router in their
topology today in order to support fast session start or ensuring
reliable connection (although with small latency overhead). At
the time ICE was designed it was not understood if this would be
too expensive or not so. ICE works without TURN but better with
it.
6. The asymmetric nature of the controlling / controlled roles has
caused many interoperability problems and bugs. Also Role
Jennings & Nandakumar Expires September 3, 2018 [Page 3]
Internet-Draft snowflake March 2018
conflicts might lead to degrade connection setup depending on
which side gets the the controlling role.
7. Priorities are complicated in dual stack world and ICE is brittle
to changes in this part of the algorithm. Although there are
advises in [I-D.ietf-ice-dualstack-fairness] specification that
might help here.
4. Snowflake for connectivity establishment
Snowflake is a light weight, asymmetric, flexible and receiver
controlled protocol for end points to establish connectivity between
them.
The following subsections go into further details of its working.
4.1. System Components
A typical Snowflake operating model has the following components
o Sender Agent: A Software agent interested in sending data
stream(s) to a remote receiver.
o Receiver Agent: A Software agent capable of receiving data
stream(s).
o Snowflake Agent: A software agent that is expected to have a STUN
Client implementation at the minimum for gathering candidates and
performing connectivity checks. Sender/Receiver agents are
Snowflake agents as well.
o CallAgent/Backchannel: Publicly reachable Server in the cloud
accessible by both the Sender and the receiver agents, acts as
backchannel/message bus for carrying signals between the Snowflake
agents.
o STUN Server: Optional component for determining the public facing
transport address of an agent behind NAT.
o TURN Server/Media Router: Recommended component acting as media
relay between the agents. A TURN Server can also act as
backchannel in certain instantiations.
4.2. Protocol Workings
The basic principle here is, each side (Receiver Agent) is
responsible for discovering a viable path for it's incoming media.
It does so by indicating the addresses for the Sender to verify the
Jennings & Nandakumar Expires September 3, 2018 [Page 4]
Internet-Draft snowflake March 2018
connectivity. Once a viable path is established, the Sender Agent
continues to transmit the media. This process deviates from ICE by
negating the need for agent's role (controlled vs controlling),
nomination procedures (aggressive vs passive) and tightly coupled
symmetric checklists validation.
As a precursor to connectivity establishment, the protocol assumes
that there exists a dedicated backchannel that the agents can use to
exchange protocol control messages.
The protocol starts with the Sender Agent conveying its intention to
send media via the backchannel to the Receiver agent. The sender
does so by sending a "PlaceCall" control message and populates the
same with the ICE candidates gathered so far.
On receiving the sender's intention to send media (via the
backchannel), the Receiver Agent proceeds with gathering the
candidates defined by its local policies or previous knowledge of
connectivity checks. The Receiver Agent then directs the Sender
Agent to carryout STUN connectivity checks towards the receiver by
sending the "DoPing" control message via the backchannel. This
message is populated with the candidate pair that the receiver wants
the sender to verify the reachability.
The Receiver Agent may sends multiple "DoPing" messages to the Sender
Agent, sending "DoPing" message per candidate pair to be tested for
connectivity, as deemed necessary. The order, the timing and the
number of candidate pairs to be tested are fully controlled by the
Receiver Agent's implementation.
On receiving the "DoPing" message with the candidate pair to be
tested, the Sender Agent carries out STUN ping checks on that
candidate pair. It does so by sending the STUN Binding Request
message towards the receiver over the media path (as its done with
ICE today). This opens up the required local pinholes and are
further maintained by the Sender for the duration of the session.
The Sender Agent also ensures that the frequency and the timing of
these checks respect the congestion control requirements for the
underlying transport.
On receiving the STUN Ping from the Sender Agent, the Receiver Agent
does the following two things:
1. It responds to the connectivity check on the media path by
sending a STUN Binding Response.
2. It also sends a "GotPing" control message with the details from
the STUN Binding Response over the backchannel to the Sender
Jennings & Nandakumar Expires September 3, 2018 [Page 5]
Internet-Draft snowflake March 2018
Agent. This is done so that the Sender Agent can verify the
connectivity status results over the backchannel as well. This
mechanism is especially beneficial for one-sided media scenarios
where the Receiver Agent can't send the STUN response to the
sender or if the response to STUN connectivity response was lost
in transmission.
If a successful STUN Ping response was received (either via the media
path or the backchannel), there is a viable path for the Sender to
transmit the media.
The above set of procedures can be continuously performed during the
lifetime of the session as and when the Receiver Agent determines
better candidates for receiving the media. Such a decision is
totally defined by the local policies and can be performed
independently of the other side.
Below picture captures one instance of protocol exchange where the
Receiver Agent indicates the Sender Agent to carry out the
connectivity checks. One can envision multiple executions of the
protocol as and when receiver has updated its knowledge of addresses
or priorities or bandwidth availability.
Snowflake Information Flow (One-way Media)
---------------------------------------------
Sender Agent CallAgent(backchannel) Receiver Agent
| | |
| | |
| | |
|(1) connect to backchannel |
|.................................................|
| | |
| | |
|Gather Sender Candidates| |
| | |
| | |
| | |
|(2) PlaceCall [Sender Candidates] |
|----------------------->| |
| | |
| | |
| |(3) PlaceCall [Sender Candidates]
| |----------------------->|
| | |
| | |
| | |Gather Receiver Candidates
| | |
Jennings & Nandakumar Expires September 3, 2018 [Page 6]
Internet-Draft snowflake March 2018
| | |
| | |
| |(4) DoPing [Candidate Pair]
| |<-----------------------|
| | |
| | |
|(5) DoPing [Candidate Pair] |
|<-----------------------| |
| | |
| | |
|(6) STUN Ping (over media path) |
|<----------------------------------------------->|
| | |
| | |
| |(7) GotPing (STUN Ping Response)
| |<-----------------------|
| | |
| | |
|(8) GotPing (STUN Ping Respnse) |
|<-----------------------| |
| | |
| | |Repeat Steps 4-8 as needed
| | |for other candidate pairs
| | |
| | |
| | |
|(9) Found a viable path for sending media |
|.................................................|
| | |
| | |
4.3. Advantages of Snowflake
4.3.1. Diagnostics
This makes it very easy to see which outbound connection were sent
from the Sender Agent to open a pin hole. Then when the Sender asked
the Receiver Agent to send a test STUN Ping, the connectivity can be
easily verified. It becomes easier to set up a client with an
automated test jig that tests all the combinations and makes sure
they work as you only need to test receiving capability and sender
capability independently.
4.3.2. Timing
This more or less removes the timing complexity by allowing both
sides to be responsible for their own timing. If it turns out that
Jennings & Nandakumar Expires September 3, 2018 [Page 7]
Internet-Draft snowflake March 2018
we can pace things much faster than 50ms then this allows us to take
advantage of that without both sides upgrading at the same time.
If we end up with a lot more candidates due to v6, mobile etc, this
removes the issue we have today where a path might have worked but
the two sides did not find it due to timing issues.
4.3.3. Asymmetric Media
This allows media to be sent in one direction over a path that does
not work in the reverse direction.
4.3.4. Fast Start
Given there exists a dedicated backchannel, this protocol can speed
up the media flow by using TURN server for the backchannel, for
example. Once either agents learns more about the candidates, each
can update the other side to ensure a better low latency path is used
for media.
5. IANA Consideration
TODO
6. Security Considerations
TODO
7. Acknowledgements
TODO
8. References
8.1. Normative References
[RFC2119] Bradner, S., "Key words for use in RFCs to Indicate
Requirement Levels", BCP 14, RFC 2119,
DOI 10.17487/RFC2119, March 1997, <https://www.rfc-
editor.org/info/rfc2119>.
8.2. Informative References
[I-D.ietf-ice-dualstack-fairness]
Martinsen, P., Reddy, T., and P. Patil, "ICE Multihomed
and IPv4/IPv6 Dual Stack Guidelines", draft-ietf-ice-
dualstack-fairness-07 (work in progress), November 2016.
Jennings & Nandakumar Expires September 3, 2018 [Page 8]
Internet-Draft snowflake March 2018
[I-D.ietf-ice-rfc5245bis]
Keranen, A., Holmberg, C., and J. Rosenberg, "Interactive
Connectivity Establishment (ICE): A Protocol for Network
Address Translator (NAT) Traversal", draft-ietf-ice-
rfc5245bis-18 (work in progress), February 2018.
[I-D.ietf-ice-trickle]
Ivov, E., Rescorla, E., Uberti, J., and P. Saint-Andre,
"Trickle ICE: Incremental Provisioning of Candidates for
the Interactive Connectivity Establishment (ICE)
Protocol", draft-ietf-ice-trickle-17 (work in progress),
February 2018.
Authors' Addresses
Cullen Jennings
Cisco
Email: fluffy@iii.ca
Suhas Nandakumar
Cisco
Email: snandaku@cisco.com
Jennings & Nandakumar Expires September 3, 2018 [Page 9]
