NAT traversal: How does ICE work?

Translation of the original in Catalan. Read it in: CatalàCastellano

Part 3 of 3NAT traversal
  1. NAT traversal: Introduction
  2. NAT traversal: How does STUN work?
  3. NAT traversal: How does ICE work?

In order to understand how

ICE

works the best thing is to explain it together with another protocol; since as was mentioned in the introduction to NAT traversal the

ICE

protocol uses a combination of the TURN and STUN protocols depending on the case, that's very simple to understand if we talk about a practical case of use. That is, when we try for example, for SIP to cross a firewall/gateway/router to deliver a call to an IP phone.

What are candidates?

The Caller sends a

SIP INVITE

message with SDP information that describes over which IP and over which PORTS the application can receive audio and/or video. These addresses and ports are known as

candidates

. These

candidates

are obtained from the

ICE

server (it's often found inside the firewall/router/gateway).

We could describe a

candidate

as an IP and a port through which a

peer

can receive data from another

peer

. There are three types of

candidates

:

  • Local candidate: local IP of the client.
  • Reflexive or STUN candidates: a public IP address of the NAT server.
  • Relay or TURN candidate: an address or a relay server that has been reserved by the client.

Traffic can always be sent successfully through the

relay candidates

, unless a firewall blocks the link, in which case there's no possible NAT traversal technique to apply. The problem of using

relay candidates

is that it requires using a server's resources, that makes the traffic arriving at the peer have delays, losses and

jitter

problems.

Process of establishing a SIP call via ICE

1. The Caller gets its candidates

In the following graph you can see how the Caller determines the

reflexive i relay candidates

for a connection. The client sends an

ALLOCATE

request to the server, then the server locates an IP/port pair (the

relay candidate

). This information is sent back to the Caller (

reflexive candidate

), this message contains the public IP the Caller will use when it emits a message to the internet, often that is the IP through which the NAT will be applied.

stun_turn_candidates

2. The Caller sends a SIP INVITE

The Caller builds a

SIP INVITE

message using the information of the candidates it has obtained from its

ICE

server, then sends it to its destination.

3. The Callee gets the candidates

When the

SIP INVITE

message is received, the Callee gets the

candidates

in the same way the Caller has done in point (1).

4. The Callee sends a 1xx response

Continuing with the SIP protocol, the Callee has to generate a response message to the

SIP INVITE

, for example, a

SIP 183

(Session Progress). When this message is built inside the SDP information the Callee will also encapsulate its

candidates

. Until this message hasn't been received and confirmed by the Caller (

200 OK

) the negotiation won't be considered successful.

5. ICE connectivity checks

Having reached this point both the Caller and the Callee know each other's candidates and it's the moment to use what is called

ICE

connectivity checks. This is done by sending a STUN message from the local candidate towards the remote candidate, this can be done for as many

candidates

as there are from the one with the highest priority to the one with the lowest. Both sides of the link will do this if both are behind a gateway with

ICE

. When it has been determined which of the

candidates

is the best to establish each of the channels then this candidate will be chosen to start sending and receiving media through it.

Below you can see three examples of how these tests of the

candidates

are done. The first diagram would be the case in which both hosts are directly routable because, for example, they're on the same LAN. The second diagram would achieve the link via the STUN methodology and in the third via the TURN methodology.

ICE_connection_check

6. The Callee sends SIP 180 (ringing)

At this point the Callee sends the

SIP 180

(ringing) message to the Caller to inform it that the Callee's phone is ringing. Having previously negotiated the data channel through

ICE

when the Callee picks up the phone it will be able to talk without delays, if this were negotiated afterwards there would be an unpleasant delay effect for the

peers

.

7. Accepts the call and they talk

So then, when the Callee picks up a

200 OK

message will be sent and via RTP (for example) and using the channels negotiated via

ICE

the multimedia data flow (in this case audio) will start crossing transparently the two gateways that give internet access to the two

peers

.

Some notes to clarify the subject

In this example the aim is to demonstrate that we can establish an

RTP

link through two

peers

that are behind NAT thanks to the

ICE

protocol. As a protocol to control the RTP session SIP is used, that makes both peers need a DNAT redirection on their gateway towards the TCP/UPD 5060 ports. In case we want to avoid that what should be used is another signalling protocol different from SIP. For example, if we don't want/can't redirect any port on the gateway XMPP could be used.

Glossary

  • Caller: who makes the call
  • Calee: who receives the call
  • SDP: Session Description Protocol
  • SIP: Session Initiation Protocol
  • Peer: one end of the connection

References