Active Topology Probing
You need:
Python, pip sudo apt-get install python, python-pip
Dotnet https://dot.net/v1/dotnet-install.sh
matplotlib sudo pip install matplotlib
numpy sudo pip install numpy
First, create a file named hosts that contains all the IPs of the machines you want to probe.
touch hosts
Remember to add one additional line to the hosts file, so some of the linux distros can count lines correctly.
One example of hosts file is:
192.168.1.2
192.168.1.3
192.168.1.4
<--- empty line
Then, you need to clone the ATP repo on all machines. To facilitate this, once you have hosts file ready, you can use MultiIssue.sh to do so.
In the ATP folder, type `bash MultiIssue.sh "git clone https://github.com/luo-liang/ATP.git".
There are three different tests to run.
Just run bash runall-ping.sh <output_name>. This does a sequential, pairwise ping of all machines. To change how long each probe runs,
emacs runall-ping.sh, then change the line cat hosts | xargs -Iip sh run-ping-on.sh ip 1 $baseName, where 1 says run each pairwise probe for 1 sec. The frequency of ping can be changed in ping-csv.sh.
Just run bash runall-tcp-seq.sh <output_name>. This does a sequential, pairwise bandwidth probe of all machines. Similarly, change ip 60 in
runall-tcp-seq.sh to change to run longer test per probe.
Just run bash runall-tcp.sh <output_name>. This does a concurrent, pairwise bandwidth probe of all machines, so each machine will establish N
connections to other machines. Similarly, change ip 60 in runall-tcp.sh to change to run longer test per probe.
Someone please do it. Should just be a minor tweak to sequential ping
Once the file is produced by each of the machines, use the raw2table folder for analysis.
cd raw2table; dotnet run <path_to_output_file> <path_to_hosts_file>
raw2table will create a folder named <output_FLDR> that contains summary. It also produces an average/min/max summary matrix, depending on the metadata of
each test, in the ATP folder, named .proced.csv. To change how raw2table produces data, for exmaple, go to each runall-*.sh command
and change the metadata banner. An example is given for runall-ping.sh
BENCHMARK:PING;SELF_TEST_OPTION:TRUE;DIMENSION:${totalClients};VALUE:MIN;PREPROCESS:0
BENCHMARK:
SELF_TEST_OPTION: TRUE so that a self to self test is performed. FALSE to avoid self to self test.
DIMENSION: number of hosts involved
VALUE: MIN/AVG/MAX, the sumamry matrix should use the minimum/average/max reading of all collected data
PREPROCESS: percentage of data to drop from beginning and end of each test, due to warmup problems and tcp slow start.
You do not need to change these metadata directives.
Use plot folder.
Run python plot.py .., and it will create time/frequency domain representation of each
546E
test you just probed. To switch between time series
and frequency series, go to plot.py and change line 9 graphType between time or histogram.