Experiment Properties

If you have extra time, you can try to change experiment properties: try different window sizes, buffer sizes, etc.

To run an experiment script with different properties, e.g. window size:

omf exec wmxtutor-streamload2.rb -- --window 5

Try running this experiment with different properties. Don't forget to reset the attenuation between experiments:

wget -qO- ""  | xml_pp

Push each experiment to iRODS and evaluate the results (Try plotting chunks on the x-axis, layers on the y-axis, and group by oml_sender_id to see how each client behaves over time).

Under what settings does streamloading outperform streaming in average number of layers played? Try creating the following plot, for each experiment result: Plot 'streamload_play' with 'oml_sender_id' on the x-axis and 'layers' on the y-axis. This will show you the average number of layers played for each client.

You can also calculate these numbers directly by running SQL commands on the result database; try (using your own expID):

sqlite3 /var/lib/oml2/expID.sq3 'select oml_sender_id,avg(layers) from streamload_play group by oml_sender_id;'

Try running the experiment a few times. Do the results change between different experiment runs? By how much? (In our tests, streamloading had a very small, but consistent advantage over streaming with respect to average number of layers downloaded per video segment).


Last modified 9 years ago Last modified on 07/22/13 22:00:30