[Lanforge] Jitter

[Ben Greear]

Rasmus Bauck wrote:
> Hmm, I agree with most of what you are saying, but if we apply jitter to
> 100% of the packages, shouldn't they then all be delayed by a RANDOM
> amount of ms (between 1 and max_jitter), and not all delayed by the
> max_jitter? The behavior we want is similar to what you are describing
> for your suggested settings, only we want it to apply to all packages.
> Unfortunately, I haven't yet had time to do much more testing on whether
> the results are more reasonable with your suggested settings, I'll get
> back to it when I have time.
>
> Another question, what distribution is used for the random jittering?
> Uniform, normal, or any other distribution?
>   
The reason 100% jitter tends to maximum latency is that packets are not 
reordered
due to jitter, so a slow packet will also delay any packets right behind 
it in the packet
stream.  If you are doing ping, or something else with low frequency 
packets, 100%
jitter will appear to work, but for high speed packet flow, you will see 
this delay
issue.

For the same reason, jittering 10% of the packets will actually make 
more than 10%
of the packets arrive at non-baseline latency times.  My suggestion is 
to use ethereal
or a similar packet sniffer and sniff traffic outside of LANforge and 
explicitly
verify that you are seeing a nice spread of jitter.

As for the distribution, it is uniform, but because of delay of 
subsequent packets
described above, the resulting jitter will tend to be more of a bell 
curve.  I do not know if there
is a particular math term to describe it..but it appears to be quite 
near what the
real world would do....

Thanks,
Ben

-- 
Ben Greear <greearb at candelatech.com> 
Candela Technologies Inc  http://www.candelatech.com