[smokeping-users] How suited is Smokeping for VoIP purposes?
nik at mitev.eu
Mon Oct 24 12:28:32 CEST 2011
I did a fair amount of experimenting and research while looking for a
monitoring solution for jitter/delay/loss and what I found in relation
to Smokeping created a couple of questions.
1. Some of the results produced seem to be inconsistent with the state
of the network and parallel probes disagree with each other.
I set up monitoring with a probe configured like
packetsize = 84
hostinterval = 0.02
mininterval = 0.001
pings = 500
offset = 0%
iptables -t mangle -A POSTROUTING -s x.x.x.x -p icmp -m length --length
112 -j TOS --set-tos 0x90
on the smokeping server to provide QoS marking (one way at least)
This produced a flat graph with hardly any delay/jitter.
When I added same probe to two more hosts, jitter and delay appeared
(3-4ms) on all hosts for the 20ms probe. Thinking that it is an issue
where FPing handles the load badly, I added one more graph to a fourth
host and expected an increase in jitter, but it went away completely and
returned to the original state for all four hosts. I am able to
reproduce that. At the same time I had a parallel probe running which
only differed in the offset - it was completely unaffected and flat at
2. Separate scaling issue follows from FPing being limited to sending
pings at 1ms interval as its highest speed. For the purposes of VoIP,
jitter needs to be measured on packets sent at 20ms intervals, which
limits the number of hosts we can monitor to 20 at a particular offset
and to 200 if we only monitor 10% of the time and split the hosts in 10
groups. Those numbers are the theoretical maximum and are likely to
actually be lower.
I am new to the list and not sure if I can attach the config files and
images to the email - I will of course email them to anyone who might be
More information about the smokeping-users