Tipping the Scales: Measuring and Scaling Asterisk

Tipping the Scales

Measuring and Scaling Asterisk

CAN YOU SPEAK MAGIC?

2

Ben Klang


3

“But…Does it Scale?!”


4

Yes!*

*Probably. This is what we’re gonna talk about


AGENDA: TIPPING THE SCALES

•Scaling terminology•Tools of the Trade•Testing Methodology•Results

5


6

Scaling Terminology


7

Calls Per Second (CPS)

Concurrent Calls (CC)


CALLS PER SECOND•Rate at which new calls are started•5CPS = 5 calls started each second•Usually meaningful only where:•Average call duration is very short (dialers)•No media is handled (proxies)•For most people, this is not the bottleneck

8


CONCURRENT CALLS•Number of calls active at any particular moment in time•This is what most people really care about when scaling Asterisk

9


HOW CPS AND CC RELATECPS * Avg. Call Duration = CC (peak)

5 CPS * 30s = 150 CC10 CPS * 30s = 300 CC

10 CPS * 60s = 600 CC

10


11

Tools of the Trade


SIPP

•THE Open Source SIP load generating tool•High Performance•Very Flexible•XML-based configuration•Awkward command line options•Very Flexible

12


13

The Problem with SIPp


STEP 1: WRITE SIP

14


STEP 2: RECORD A CALL

15


STEP 3: DECIDE SIPP ARGS

16

sudo /usr/local/sbin/sipp -p 8836 \ -sf scenario.xml -l 1000 -m 20000 \ -fd 1 -r 5 -s 1 \ -no_rate_quit -rate_max 35\ -rate_increase 1 -rate_interval 30 \ -trace_stat -stf stats_file.txt\ -i 10.3.18.108 -mp 20953 -trace_screen \ -screen_file screen_file.txt \ -trace_err -error_file errors.log \ -t u1 10.3.18.134


STEP 4: TEST

17


PERMUTATIONS!

18


19

The Problem with SIPp:


20


SippyCup

1. Write test steps in simple syntax

2. Compile SIPp XML + PCAP audio

3. Run SIPp with correct arguments

21


EXAMPLE: SIPPY CUP

22

source: 192.0.2.15destination: 192.0.2.200max_concurrent: 1000calls_per_second: 5number_of_calls: 20000steps: - invite - wait_for_answer - sleep 3 - send_digits '4045551234' - sleep 5 - send_digits '#' - wait_for_hangup


TIME TO CREATE ONE TEST

23

SIPp: ~1 hour

SippyCup: ~1 minute


24

SIP Treadmill


25


26


27

Testing Methodology


BAD QUESTIONS•How many calls can Asterisk handle?•I heard “X” is faster than “Y”?•How much CPU/Memory do I need?•Does it scale?!

28


GOOD QUESTIONS•What is my average call duration?•Am I touching the media?•What if I enable recording?•Can I optimize out transcoding?•Do I need mixing/conferencing?

29


30

How much does each Asterisk feature

impact the capacity of a given system?


THE PERFORMANCE QUADUMVIRATE

31


32

Results


BASELINE: B2BUA

33


BASELINE: B2BUA

33


BASELINE: B2BUA

33

Saturation Point


FILE PLAYBACK (NATIVE)

34



34



34


FILE PLAYBACK (TRANSCODED)

35



35



35


CONFERENCES (~10 CALLS/ROOM)

36



36



36


RECORDED CALL

37


RECORDED CALL

37


RECORDED CALL

37


THE RESULTS

38

BaselinePlayback (native)

Playback (xcode)Split Conf

Recorded

49%

66%71%75%

100%


IF YOU HEAR NOTHING ELSE…•Test for yourself!•Open Source tools make this not as hard as it used to be•Sippy Cup•SIP Treadmill•SIPp

39


40

@MojoLingo

@bklang [email protected]

sipp.sf.net

mojolingo.github.io/sippy_cup

github.com/att-innovate/SIPTreadmill

Tipping the Scales: Measuring and Scaling Asterisk

Technology