[asterisk-users] [OT] Load testing with SIPp

[Bruce Ferrell]

On 03/06/2018 05:47 PM, Olivier wrote:
> All mentioned boxes are VMWare VMs probably configured with default drivers and settings.
> Unfortunately, I don't know which network equipements are used when those boxes communicate with each other.
> So yes, maybe I banged into an unexpected limit because of this ignorance.
>
> Anyway, from my testing, I was mostly surprised  by 2 things:
>
> 1- I consistently hit a 200 calls when I let run an Asterisk instance on the SIPp (calls passing locally from SIPp to a first Asterisk instance before hitting SUT): I thought 
> that communications inside a given VM was "unlimited and cheap" and I was apparently wrong,
> Disabling Asterisk in my first VM allowed me to hit a 500 limit at the price of loosing Asterisk flexibility to log RTCP stats, select exotic codec and so on.
>

What you've banged into is called "over subscription".  NOTHING is unlimited and "cheap"  There Ain't No Such Thing As A Free Lunch.

Virtualization works best when the guest VMs aren't 100% duty cycle... When one guest is idle, another can use the resources of the host.  When the guest VMs all present a 
significant load in any area, the host WILL bog down and so will all of the guest systems.

You might be better of not running this as a virtual app.

> 2- Increasing asterisk.conf's maxfiles value (from 10000 to 40000, for instance), allowed me to get more than 695 successfull call out of 700 but I could succeed to get, even 
> once, 700 successfull calls, even when I tried with my current 400000 maxfiles limit.

See the above... While you tweaked maxfiles on the guest, the host came came into play.

> 2018-03-06 23:35 GMT+01:00 Bruce Ferrell <bferrell at baywinds.org <mailto:bferrell at baywinds.org>>:
>
>
>     On 03/06/2018 01:58 PM, Olivier wrote:
>
>         Hello,
>
>         I'm running load testing sessions.
>         My System Under Test is an asterisk 13 with 16GB, configured with maxfiles set to 400 000.
>         This system is supposed do produce simple SIP trunking services without transcoding.
>
>
>         The box sending call to my System Under Test is anabled with SIPp.
>         I'm banging on a 700 concurrent calls/50 CAPS limit I would like to improve, if possible.
>
>         Tests are done with both signaling and media like this:
>
>         SIPp <---> SUT (asterisk 13) <---> Asterisk box echoing media
>
>         I checked bandwidth first and got 930 Mb/s on each leg (from SIPp to SUT or SUT to echoing box) using iperf3 TCP testing though my target relies on UDP
>
>
>         My questions are:
>
>         1. Have you ever noticed a better scalability using UDP or TCP ?
>
>         2. Where do Retransmission I'm observing on SIPp console most probably come from ? Network issues ? My SIPp not beeing correctly tuned ? Lack of resources somewhere ?
>
>         3. Recommandations ? Suggestions ?
>
>         Best
>
>
>     I do network management for a living.
>
>     In your description, I see nothing to describe the network other than an observed 930Mb/s.
>
>     What is the network configuration;  What NIC(s), switches etc.
>
>     Treating these as effectively "unlimited" is a certain recipe for banging into unexpected limits.
>
>     Different NICs and switchs can and do provide differing levels of performance.
>
>