Linux H.323 setup and configuration

What H.323 is in theory and in practice.

H.323 is a series of related standards published by the ITU, and is largely based on the ISO family of standards.

It defines how one endpoint may discover another, call it, and transfer data.

An endpoint can be either a process on a computer or a real telephone; it can be identified with a telephone number in both cases, or with an IP address if it is a process running on a computer.

Most endpoints are terminals, or they can be gateways in which case their main role is to act as intermediaries between other endpoints.

A gatekeeper is a special type of endpoint to which other endpoints can connect to register a symbolic name or number for themselves, and usually to ask the gatekeeper to make or receive calls on their behalf. Somewhat similar to a PBX/switch in function.

A gatekeeper is usually required for calls to real telephones, in which case the gatekeeper acts as gateway. When both endpoints involved in a call with a gatekeeper in the middle are IP based, it operates as a proxy, and this for example is useful when NAT/masquerading is involved. Gatekeepers can be either peers or endpoints to other gatekeepers.

H.323 mostly applies to calls between endpoints where at least one of them is a process with an IP address. It might also be used for calls between real telephones, where at least a segment of the path (usually the longest ins some sense) is over the Internet.

These notes however are only for the case where all endpoints involved are processes with IP addresses.

In practice H.323 exists in two variants: the original one, and the somewhat modified variants that Microsoft has introduced with NetMeeting. Even if NetMeeting has been retired by Microsoft, it is still somewhat widely used. The NetMeeting variant contains some incompatible extensions or restrictions.

However, there are at least a couple of simple, small but functional freeware MS Windows applications that can be used instead of NetMeeting, and thus these notes will not mention the NetMeeting variant much, as it is possible to use the freeware alternatives instead.

Some details about the H.323 procotols

First of all, the data (sound/video/...) transport protocols of H.323 are trivial; it's just RTP over UDP, and they sort of just work.

All the major difficulties arise out of endpoint discovery and call setup, especially in the complicated cases where one has firewalls, gatekeepers, and reflectors.

Setup instruction, configuration files and optiona often refer by name to the invidual standards and protocol that are part of H.323, for example:

H.225.0 RAS

UDP protocol, unicast or multicast, for interacting with a called endpoint which is a gatekeeper, and includes asking the gatekeeper for registration, for calling another endpoint, etc.

H.225 non-RAS

TCP protocol for interacting with a called endpoint which is a terminal.

Q.931

TCP based procotol, used embedded in H.225 non RAS, for setting up direct calls to endpoint which is a terminal.

H.235

Protocol embedded inside H.225 for secure authentication of an endpoint to a gatekeeper. Note that this is not symmetrical; gatekeepers never authenticate themselves to their endpoints.

H.245

TCP, complementary to H.225, can be embedded within it, deals with setting up data connections using RTP over UDP, text messaging etc.

H.450

Call rediction of various sorts (hol,d, forwarding, transfer).

RTP

Purely UDP, is used to carry real time streams of voice or video data.

Within H.225 there are a number of significant interactions between an endpooint (or a peer) and a gatekeeper:

Q.931 Setup

Call setup from endpoint directluy to another endpoint. Usually to port 1720 of the called endpoint, and from port 1720 of the caller.

H.225 RAS

The various interactions between an endpoint and the gatekeeper it uses (Registration, Admission and Security):

• GRQ: Gatekeeper ReQuest (connect to gatekeeper)
• RRQ: Registration ReQuest (log into gatekeeper)
• IRQ: Information ReQuest (gatekeeper queries endpoint)
• ARQ: Admission ReQuest (call another endpoint via gatekeeper)
• LRQ: Location ReQuest (ask another gatekeeper about endpoint)
• BRQ: Bandwidth ReQuest (ask gatekeeper for bandwidth change)
• DRQ: Disengage ReQuest (close call to another endpoint)
• URQ: Unregister ReQuest (log out of gatekeeper)

Random notes to be sorted out

• It may be important to improve the latency of sound under Linux as illustrated in these notes.
• GNUGK: Only the first alias is ever tested for authentication, even if all are valid as call destinations.
• GNUGK: in the [Password] section the authentication tokens on the right hand side include the alias name, so if one wants to create multiple entries with the same password one cannot just copy the authentication token.
• openh323: the LPC-10 codecs seems badly broken, as it causes crashes with GnomeMeeting or strange loops with ohphone.
• openh323: theoretical codec bandwidth numbers are somewhat inaccurate. In practice I have observed the following:
  • SpeexNarrow 8kb/s: 3000 bytes/second (about 24kb/s).
  • SpeexNarrow 15kb/s: 3500 bytes/second (about 28kb/s).
  • MS-GSM 13kb/s: 2700 bytes/second (about 22kb/s).
  • GSM-06.10 13kb/s: 2200 bytes/second (about 16kb/s).
  In practice only GSM-06.10 has a low enough bandwidth to work well on the uplink of a 56k modem.
  There is a more extensive comparison of codecs at UNINETT which illustrates that details of the packetizing have a large influence.
• openh323: in general, in .ini files, multiple lines with the same string on the left hand side result in a single entry with that string as key and a value which is the concatenation of the right hand sides.
• OpenPhone: it has a bug by which it pads with zeroes aliases to a minimum of 3 bytes, for example:

      destinationInfo = 1 entries {
        [0]=h323_ID  3 characters {
          0070 0067 0000                            pg 
        }

  which of course means that aliases of 1 and 2 bytes are not going to work with a correctly functioning gatekeeper like GNUGK.
• GnomeMeeting: It is possible to put spaces in h323: URIs, but they get misprinted in the status line.
• GnomeMeeting: which alias gets selected as the first one? Complex rules:
  • If both First name and Surname are specified, the alias is their concatenation separated by a space; if Gatekeeper alias is specified, that will be used as the second alias.
  • If only one of First name or Surname is defined, then that is the first alias, and then the Gatekeeper alias is used as the second alias, unless it is the same string as the first, in which case there is no second alias.
  • If neither First name nor Surname are defined, then the login name is first alias, and then the Gatekeeper alias is used as the second alias, unless it is the same string as the first, in which case there is no second alias.
• ohphone: it is possible to specify multiple aliases by repeating the --username option.
• openmcu: if the --user option is repeated, this does not create multiple aliases, but a single alias that is the concatenation of the various values specified, with spaces in between.
• ohphone, openmcu, GnomeMeeting: it may be very helpful to run these at a real time priority. To avoid running them as root, start them as a normal users, then change their priority class to FIFO using chrt, for example like this:

  for p in $(pidof ohphone); do chrt -v -r -p 5 $p; done