BLOG     |     FORUM
Asterisk Call Analyzer for IAX2
An add-on tool for complete IAX2 (Asterisk Protocol) call analysis
  Click for Whitepaper "IAX2 Call Analyzer for Unsniff" (900 kb)IAX2 Call Analyzer
Are you one of the growing number of people deploying the Asterisk VoIP platform ? We present a add-on tool that performs complete VoIP Call Analysis for the Inter Asterisk Exchange (IAX2) protocol. Measure and plot call bandwidth, interarrival delay, jitter, packet loss, and IAX2 events for each direction of call. This tool like others in this series (such as TCP/IP analysis) is written in the excellent Ruby scripting language using the Fox-Ruby toolkit. Full source code of the tool is provided for you to tweak it to your liking. If you are working with Asterisk in any capacity, this is a "must  have" tool on your workbench.

IAX2 Call Analyzer  

Calculate and plot bandwidth, delay, jitter

Asterisk (the open source PBX server) is rapidly gaining in popularity as a powerful alternative to expensive PBX systems. The IAX2 (Inter Asterisk Exchange ver 2) protocol is the native language of Asterisk. The main strength of IAX2 when compared to competing protocols such as RTP/SIP/H.323 is its friendliness to NAT (Network Address Translation) and firewalls. IAX2 uses only a single UDP port 4569 to carry both media and control messages.

How to download and run this tool

Asterisk Analyzer
IAX2 is a relatively new protocol and there is a lack of tools available that can perform quality analysis of IAX calls. Ethereal (the excellent open source protocol analyzer) contains IAX2 protocol dissection but lacks aggregate call analysis features.

The features of the IAX call analyzer are:

The tool is written in Ruby using the powerful Unsniff Scripting API. The tool uses the Fox-Ruby and the UnleashCharts charting library to build its user interface.The tool comes with full source code. You can modify / adapt / extend it any way you wish. Please let us know of any bugs in the tool or improvements you would like to see.


Extract calls from an capture file

If Unsniff sees a NEW IAX control message and a corresponding ACCEPT message, it assumes a call has been set up and creates a new call. The data associated with each call is extracted from these two messages and shown in a table.  You can then double click any call in the table to analyze that call.

Track asterisk calls
For each call the following data is shown.
From IP , To IP , Source Call Number, Dest Call NumberThese four attributes are used to uniquely identify a call.
Start TimeWhen was the call started ? This corresponds to the time when the ACCEPT message was seen.
Duration SecondsHow long was the call in progress ? This is the time difference between the IAX Control HANGUP message and the call start time.
CodecWhat codec was used for the call ?
Calling numberThe number or extension of the calling phone
Calling nameThe name of the calling phone
Called numberThe number or extension of the called phone (maybe a SIP URI)
DNIDThe dialled number ID
UsernameThe user who made the call
 


Call Bandwidth Chart

One of the most important factors that needs to be considered while deploying a VoIP system is the bandwidth requirements. The bandwidth at your disposal will help you plan tradeoffs on voice quality by selecting the right codec. Each codec has a specific voice stream bandwidth. For example G.711 generates a packetized voice stream of 64kbps bandwidth. However, the actual bandwidth required is larger due to the IP/UDP/IAX2 headers. The call bandwidth chart plots the actual bandwidth usage of the selected call.

The methodology used by the bandwidth analyzer is:
Use entire packet size including Ethernet/IP/UDP/IAX2 headers
The entire packet size is used to calculate the bandwidth requirements. As an example, if we are analyzing a call which uses the GSM codec, the total payload is calculated as: Total payload = 33 (GSM) + 4 (IAX2) + 8 (UDP) + 20 (IP) + 14 (Ethernet) = 79 bytes
You can compare your bandwidth utilization with the specifications different codecs at  http://www.openh323.org/docs/bandwidth.html

Sample used bandwidth every 200 ms
We use a 200ms sample rate of used bandwidth to plot. If you wish to use a more granular sample, then you can change the line @sliceus = 200000 line in the iax2ana.rb script.

In the picture below we can see that the call uses the iLBC codec and the used bandwidth is around 35kbps in each direction. You can play with this tool for other codecs such as G.711, GSM,  Speex and verify for yourself the bandwidth usage claims of each of these codecs.
ilBC Codec Call Bandwidth in reverse direction


One-way delay (Latency)

When you talk into a microphone,the first thing that happens is the sampling of your voice (often at 8000 times per second). The one-way delay is the time between the sampling of a voice signal and the playout of the sample at the other end. This includes codec delays, network buffering, transmission delays, and playout buffering. While all these factors are important to determining the overall voice quality, the transmission delay is the largest and most unpredictable component. Even casual users of voice over IP services would have noticed that as latency gets longer, it becomes much harder to carry out a meaningful conversation. The well accepted benchmark for maximum one-way delay is 150ms. Unfortunately, measuring one-way delay is not easy. We need timestamp synchronization between the two endpoints and a way to carry this information in each packet. This is too much to ask in a loosely controlled network such as the internet. So what we are really saying is, Unsniff cannot calculate the one-way delay for given these requirements. Luckily it turns out the inter-arrival delay (next section) is a pretty good indicator as well of voice quality. Unsniff can calculate the inter-arrival delay and plot them on a chart for the duration of the call. So lets accept that one-way delay calculation will remain an elusive goal for us and move on to the next section !


Interarrival Delay

In the previous section, we saw that measuring one-way delay is not possible given our constraints.  Let us now focus on the next best thing, the inter-arrival delay.

What is Interarrival delay ?
Assume a VoIP transmitter is sending voice packets exactly 20ms apart, so they keep coming at 0ms, 20ms, 40ms, tick,tick,tick,tick, . In an ideal situation the receiver must also receive these packets exactly 20ms apart (at t+0ms, t+20ms, t+40ms) and so forth. Actually this is what happens in traditional phone systems. Unfortunately in an IP network, this not the case. While we can easily arrange the sender to transmit packets at exactly 20ms intervals, we cant "arrange" for them to be received at exactly the same rate. This is due to the routers and the packet switched network in between. So we may have a case where the receiver gets the packets at (t+0, t+23, t+45, t+61, etc). This means that instead of arriving nicely at 20ms intervals, we get packets at 23ms, 22ms, and 16ms intervals. Since the actual numbers (23ms,22ms,16ms above) are dependent on the transmit rate (20ms), we define interarrival delay as the difference between the two. In this case we have interarrival delays of 3ms for the 1st packet, 2ms for the 2nd packet, -4ms for the 3rd packet. So we have delays of 3ms, 2ms, 4ms in the above example. This is how interarrival delay is calculated. This is also sometimes misleadingly referred to as simply “delay”.


Calculating interarrival delay for IAX2Interarrival Delay Chart
The two measurement points required to calculate delay are the received time and the transmit time.
Received Time
The received time is when the packet was captured by Unsniff. Getting this is not a problem because Unsniff gets a microsecond resolution timestamp for each captured packet from its provider (Winpcap or Windows Raw Sockets). The only tricky part it to position Unsniff close to the receiver if we want a reasonably accurate measurement. Ok next..
Transmit Time
This is when a packet was transmitted. Unsniff does not have access to this information because it is only running near the receiver. You could probably run another instance of Unsniff near the sender and correlate two packets by writing simple scripts. This is not feasible in a number of situations. The IAX2 timestamp comes to our rescue. The timestamp in the IAX2 mini frame is the number of milliseconds since the call began. We can use this timestamp because we are only interested in the time difference not the actual time. All we have to do is convert the microsecond resolution timer at the receive side to milliseconds and we have our measurements.




Jitter (statistical variation of Interarrival Delay)

Within bounds Interarrival delay can be easily controlled by adding a buffer. Sometimes interarrival delays are not problem at all because they are uniform. Let us consider an example : If at a receiver packets arrive at (t+0, t+22, t+40, t+62, t+80), we can see that the delays are +2ms, -2ms, +2ms, -2ms and so on. So you can just add a buffer for a single packet and still play it out with no problems, except that you introduce some additional delay due to the buffer. The real problem happens when these interarrival delays vary. This interarrival delay variation is known as the jitter (a term borrowed from electronics).

The jitter is quite a good indicator of voice quality. However, it does not mean much at a single point of time. So, if someone walks up to you and says “My call experienced a jitter of 8.3ms at 10:42:22 AM”. You will not know what to say about the voice quality at that instant. You can use jitter to compare different calls or compare time periods of the same call. If the same person walks up to you and says, “My call experienced high jitter (between 7ms and 8ms) for a few minutes at 10:42:22AM compared to the beginning of the call (between 0ms and 2ms)”. You can conclude that the voice quality during the phase of high jitter was lower than during the period of low jitter.


Calculating jitter for IAX2Jitter Chart for IAX2
The IAX2 informational draft does not include a formula for calculating jitter from interarrival delay samples. So we stole one from RFC3550 (RTP - A transport protocol for real time applications) - the predominant VoIP protocol. We think that the formula is valid for both RTP and IAX2.

The formula is :
Jitter (at time T) = Jitter (at time T-1) + ( Interarrival Delay (at time T) - Jitter (at time T-1) ) x 1/16

So the jitter is calculated continuously over the duration of the call. You may ask , What is the mutliplier "1/16" doing up there in the formula ? It is the “gain parameter”. Here is what RFC3550 has to say about the 1/16 factor. "This algorithm is the optimal first-order estimator and the gain parameter 1/16 gives a good noise reduction ratio while maintaining a reasonable rate of convergence.". 


IAX2 Events


We have looked at call bandwidth, interarrival delay, and jitter charts for voice over the IAX2 protocol. While all these give you an idea of the quality of the call. You still want to see how the IAX2 protocol itself works over the timeperiod of a single call:
  • what messages are sent at what points in time ?
  • when was the call setup / call hung up
  • were there any DTMF digits transmitted
  • was the call transferred
  • all other IAX2 messages sent (such as PING/PONG/LAGRQ, etc), when were they sent
  • how many mini frames vs full voice frames

Let us take a look at a sample call.

IAX2 Event Chart


The notation used in the IAX Events chart is shown in the table below

White Dots along top around the 55 markIAX2 Mini Frames containing voice samples from initiator of call Yellow Dots along top around the 45 markIAX2 Mini Frames containing voice samples from receiver of call
NNEW >PING
RRegistration messages <PONG
V - yellowFull frame containing voice +ACK
# DTMF digit pressed any RED colorRetransmitted packet
AAccept LLag messages (LAGRQ  and LAGRP)
OOther IAX Control message *RINGING
@ANSWER UOther Control message




How to Download and run this tool

Pre-requisites

1. You need to have Unsniff Network Analyzer installed on your system (download a trial version from here)

2. You need to have Ruby installed on your system (click here to download a one-click installer for Windows)

Download the IAX2 Call Analyzer script

1. Download iax2ana.rb (the IAX2 Call Analyzer Ruby Script) to a folder on your PC

2. Download UnleashCharts.rb (the Unleash Charting Library) to the same folder 

How to run ?

  • Capture some IAX2 packets from the network under test 
  • Save the captured packets 
  • You can then run the IAX2 Call Analyzer on the captured file 
  • Once you bring up the IAX2 call analyzer, double click on any call to draw charts of that call

Usage:

iax2ana <capture-file-name>

capture-file-name : Capture file in Unsniff (*.usnf) format 

Example:

c:\RubyTest> iax2ana.rb  BuggyAsteriskCapture.usnf