Re: quesiton about PerfSocket_UDP.cpp
Hey Kevin,
Thanks for the information. What I am trying to figure out is why there is such a huge
variation in Jitter from one period to the next. From my previous post you may have
seen that during a 3 minute transmission I get a sawtooth pattern in Jitter delays. What I
am trying to figure out is if during one period ( where the jitter is extremely low) the packets
are arriving at a fairly consisten pattern and in another ( where the jitter is high during the
same transmission) the packets are arriving quicker followed by slower patterns generating
a higher jitter during that period. You can see my capture below.
[ 3] 52.0-52.5 sec 64.6 KBytes 1.06 Mbits/sec 1.377 ms 0/ 45 (0%)
[ 3] 52.5-53.0 sec 67.5 KBytes 1.11 Mbits/sec 3.826 ms 0/ 47 (0%)
[ 3] 53.0-53.5 sec 64.6 KBytes 1.06 Mbits/sec 1.412 ms 0/ 45 (0%)
[ 3] 53.5-54.0 sec 67.5 KBytes 1.11 Mbits/sec 3.280 ms 0/ 47 (0%)
[ 3] 54.0-54.5 sec 64.6 KBytes 1.06 Mbits/sec 1.314 ms 0/ 45 (0%)
[ 3] 54.5-55.0 sec 67.5 KBytes 1.11 Mbits/sec 4.084 ms 0/ 47 (0%)
[ 3] 55.0-55.5 sec 64.6 KBytes 1.06 Mbits/sec 1.105 ms 0/ 45 (0%)
[ 3] 55.5-56.0 sec 70.3 KBytes 1.15 Mbits/sec 3.441 ms 0/ 49 (0%)
[ 3] 56.0-56.5 sec 63.2 KBytes 1.03 Mbits/sec 1.062 ms 0/ 44 (0%)
[ 3] 56.5-57.0 sec 67.5 KBytes 1.11 Mbits/sec 3.665 ms 0/ 47 (0%)
[ 3] 58.0-58.5 sec 64.6 KBytes 1.06 Mbits/sec 1.015 ms 0/ 45 (0%)
[ 3] 58.5-59.0 sec 61.7 KBytes 1.01 Mbits/sec 0.904 ms 0/ 43 (0%)
[ 3] 59.0-59.5 sec 70.3 KBytes 1.15 Mbits/sec 1.185 ms 0/ 49 (0%)
[ 3] 59.5-60.0 sec 63.2 KBytes 1.03 Mbits/sec 1.184 ms 0/ 44 (0%)
[ 3] 60.0-60.5 sec 68.9 KBytes 1.13 Mbits/sec 1.098 ms 0/ 48 (0%)
What is odd is that the pattern alternates low jitter then higher then lower then even
higher until finally it is fairly steady for a few periods. This pattern then repeats itself.
This pattern only shows itself over a wireless connection. When I run the test over a
wired connection I get the "normal" fluctuating jitter pattern. I was just hoping
someone had seen this pattern over a wireless network and had an explanation.
I guess its the rolling jitter variance pattern that is so odd.
iperf-users --at-- dast.nlanr.net writes:
>As per the source we are using the algorithm defined in RFC 1889 for the
>Real Time Protocol to calculate jitter, defined in section 6.3.1 and
>coded in A.8 (which looks very familiar to the iperf code). That formula
>is:
>J = J + ( | D(i-1,i) | - J ) / 16
>The deltatransit is the "D(i-1,i)" part. It is measured in seconds and is
>a double type.
>
>Therefore this is not really a difference from average (since we don't
>compute an average), but a difference from the last packet. I hope that
>helps, but I am not sure it was exactly what you were looking for.
>
>Kevin
>
>
>On Wed, 31 Mar 2004, John J. Sheftic wrote:
>
>> I am trying to determine some jitter variations in my wireless network
>> and was trying to record the differences in transit time across the
>> network. Since Jitter is an unsigned value it is difficult to tell
>> exactly if packets are consistently arriving
>> late or both early and late compared to the average. In the
>> PerfSocket_UDP.cpp class I noticed a deltaTransit variable and that
>> looks promising. What exactly is deltaTransit calculating and what unit
>> is it reported in. I need to be able to see if
>> packets are arriving early and then late causing Jitter to rise or
>> whether it is a consistent late arrival. Thanks for your help.
>>
>