[smokeping-users] How suited is Smokeping for VoIP purposes?

[Nik Mitev]

Chris

On Wed, 2011-10-26 at 10:34 +0100, Chris Wilson wrote:

> I think this is a bug with FPing. Smokeping relies on fping to do the 
> actual pinging in the default configuration, although you can choose a 
> different pinger. I remember trying to debug fping to solve a different 
> bug and finding it trying, but if you program C or you can reproduce the 
> problem on the fping command line, then it's possible to report 
> upstream or investigate further.

Yes that is a very likely cause. FPing is however the only smokeping
probe I found capable of sending a stream of pings, with no
configuration required on the target. TCP/UDP pings will not work for
most targets.

> It's theoretically possible to make fping send any number of pings per 1ms 
> interval, although further skewing the time-distribution of pings may 
> exaggerate the effects of any network boundaries beyond what you'd 
> experience in a real call.

That is true and I am not considering probing one host with pings at a
rate higher than 20ms. The issue arises when we need to ping multiple
hosts at the same time at 20ms intervals. If the global limit for FPing
is 1ms and it cannot send more than 20 pings in 20ms, we can configure a
maximum of 20 hosts to be probed simultaneously.
> 
> Also beware that if you monitor continuously at this rate, your monitoring 
> is going to have a major impact on network bandwidth use and thus 
> negatively affect call quality.

I agree this could be an issue if we had to carry monitoring traffic
over the Internet. At the moment everything stays on campus so even
80Mbps for 500 probes (as a ballpark figure) should be fine as long as I
put the smokeping server on a 10G rather than 1G link.
That said I am looking at using Argus data for measuring jitter on
existing call traffic, but it is yet to be seen how reliable that is.

Cheers, Nik