...
This is a study of the differences (if any) of the measurements from pinger.slac.stanford.edu (a single (raw bare hardware non-virtual) Linux host (Dell PowerEdge 2650) running the PingER Measurement Agent (MA) measuring the ping Round Trip Times (RTTs) to and pingervm (AKA dhcp-nebula-124-66.slac.stanford.edu a Nebulr KVM Virtual Machine) and vice versaa Nebula KVM Virtual Machine with floating addresses) MA measuring the ping Round Trip Times (RTTs) to hosts worldwide and between pinger and pingervm. Hencefiorth we refer to ping requests from pinger to pingervm as pinger to pingervm or pinger>pingervm and ping requests from pingervm to pinger as pinger to pingervm or pingervm>pinger. Both pinger and pingervm are located on the second floor of Building 50 (the computer center) at SLAC. The traceroutes between the two machines are seen here.
The measurements were made from February 26th 2015 and March 3rd 2015. Every ~30 minutes a burst of 10-30 (10 if no loss else up to 30 until 10 ping responses were received) ping requests was sent by the MA to the target. The ping packet size was 100Bytes, the ping requests in the burst were separated by 1 second. The number of pings between pinger and pingervm was ~6000.
Below we look at a statistical analysis of the results for pings from pinger and pingervm to the World, to N. America, Europe and between pinger and pingervm.
To World (excluding SLAC targets).
The following tables are from pingtable.pl
...
PingER non VM measurement agent
| Tick | min | 25th% | avg | median | 75th% | 90th% | 95th% | max | iqr | std dev | # pairs |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Mar2015 | 23.312 | 178.346 | 239.329 | 211.152 | 293.499 | 339.425 | 379.947 | 780.659 | 115.153 | 104.510 | 107 |
| Feb2015 | 22.689 | 176.167 | 241.608 | 212.435 | 295.448 | 326.637 | 362.350 | 803.815 | 119.281 | 110.212 | 112 |
...
PingER VM Measurement agent
To Europe
Tick | # pairs | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Mar2015 | 151.017 | 162.965 | 176.121 | 173.904 | 185.921 | 199.620 | 208.584 | 208.584 | 22.956 | 17.586 | 14 |
To N. America
...
PingER non VM measurement agent
| Tick | min | 25th% | avg | median | 75th% | 90th% | 95th% | max | iqr | std dev | # pairs |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Mar2015 | 23.312 | . | 54.890 | 62.042 | 79.315 | 79.315 | 79.315 | 79.315 | . | 28.678 | 3 |
...
And plots of the time series appears as below (spreadsheet)?
Looking at the above manually scaled plots of pinger>pingervm and pingervm>pinger it is apparent the RTTs from pinger to pingervm are > pingervm to pinger. Looking at the average, standard deviation, median and IQR tables and the differences between pinger as the monitoring site and pingervm as the monitoring site, we see in tabular form (where the average errors are +- (stdev(pinger>pingervm)+stdev(pingervm>pinger)) and the median errors are +- (IQR(pinger>pingervm) + IQR(pingervm>pinger)). the Probabilities are those that the pinger>pingervm and pingervm>pinger distributions are the same (assuming normal distributions).
...
By looking at the cumulative distributions we can get the Median differences probability.
Minimum RTT distributions between pinger and pingervm
We focus on the minimum RTT since that is least affected by queuing etc.
