Using Perl and Linux for writing Real-Time Software
and other various hopefully-interesting things

Specifically, voice-over-IP

Quick intro to VoIP

- RTP (RFC 1889)
  - Real Time Protocol

  - take audio data, put a timestamp and sequence number on it, and
    put it in a UDP packet

- SDP (RFC 2327)
  - Session Description Protocol

    v=0
    o=Cisco-SIPUA 8522 8274 IN IP4 203.x.y.z
    s=SIP Call
    c=IN IP4 203.x.y.z
    t=0 0
    m=audio 26858 RTP/AVP 8 0 18 101
    a=rtpmap:8 PCMA/8000
    a=rtpmap:0 PCMU/8000
    a=rtpmap:18 G729/8000
    a=rtpmap:101 telephone-event/8000
    a=fmtp:101 0-15
 
- SIP (RFC 3261)
  - Session Initiation Protocol

    INVITE sip:voicemail@203.x.y.z SIP/2.0
    Via: SIP/2.0/UDP 203.x.y.z:5060
    From: "Geoffrey" <sip:g@203.x.y.z>;tag=00036b8b191e0009534ba7eb-785e52b8
    To: <sip:voicemail@203.x.y.z>
    Call-ID: 00036b8b-191e0011-70640599-5b37a907@x.y.z.87
    Date: Tue, 15 Jul 2003 07:45:29 GMT
    CSeq: 101 INVITE
    User-Agent: CSCO/4
    Contact: <sip:g@203.x.y.z:5060>
    Expires: 180
    Content-Type: application/sdp
    Content-Length: 248
    Accept: application/sdp
    Remote-Party-ID: "Geoffrey" <sip:g@203.x.y.z>;party=calling;id-type=subscriber;privacy=off;screen=no

Phone diagram

Commonly use packets with 20ms of audio = 50pps.

Real-Time
- Hard real-time: deadlines (usually <1sec) must be guaranteed or the
  system fails catastrophically
- Soft real-time: deadlines must be met most of the time

VoIP requires sending packets out with reasonable consistency.
Receiver will have a jitter buffer to deal with packets arriving late.

Implementing SIP end-points:
- Music On Hold, Voicemail, IVR, Conference Bridge

Is Perl suitable?

time() accurate to 1 sec
sleep() accurate to 1 sec

gettimeofday() accurate to microseconds
select() can specify microsecond durations

What happens?  see test-select, test-select2

Why?  Linux kernel 100Hz timer.
Can patch for higher values
(interesting sidenote -- HZ patch for boot-time selection, tickless patch)

RTP packing/unpacking

    0                   1                   2                   3
    0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |V=2|P|X|  CC   |M|     PT      |       sequence number         |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                           timestamp                           |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |           synchronization source (SSRC) identifier            |
   +=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+ 
   |            contributing source (CSRC) identifiers             |
   |                             ....                              |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                           payload                             |
   |                             ....                              |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

pack()/unpack() in Perl?

Inline.pm
- embed C in Perl
- write as little C as possible
  - enough to convert to and from a binary string and a Perl hash:

    my $packet = {
      data         => '...',
      marker       => 1,
      payload_type => 8,
      seq_num      => int(rand(65536)),
      ssrc         => int(rand(2**32)),
      timestamp    => int(rand(2**32)),
    };

First experiments didn't require coding or decoding of audio, so
non-issue at the time.

- ethereal

Architecture to handle more than one thing at a time:

Three layers:

- SIP Messaging - should be responsive, but not time critical
- Application Logic - can pause for long periods of time
  (Separate programs for different functionality;
  one instance per call)
- RTP (One instance per call) - time-critical

RTP layer has two functions
- Tx/Rx audio from the network
- Read/write audio from the disk

Threads?  One for disk I/O, one for network I/O?

  use Thread;
  use Thread::shared;
  share(..);
  my $t = Thread->new(\&thing_to_start, 'arg1', 'arg2');

Shared scalars leak :(

Codecs & File I/O
- Inline again
- libsndfile

Debugging
- Perl debugger
- gdb on Perl for Inline
  use Inline C => 'DATA',
             OPTIMIZE => '-g',
             FORCE_BUILD => 1,
             CLEAN_AFTER_BUILD => 0;
  % gdb perl
  gdb> set args -d rtp
  gdb> r
  perl> n
  perl> n
  perl> ^C
  gdb> b pack_rtp
  gdb> l
  gdb> c
  perl> c 44
  perl> n
  perl> x pack_rtp($packet)
  gdb> ...