Analysis conclusion

The measurements taken, although anticipated, allowed us to understand with details the behavior of a streaming system, with one server and multiple clients. All three parameters we measured were correlated in the experiments of both scenarios. Also as expected, the observable differences between the two scenario, hence the two modes, were particularly marked on the CPU load and the Storage used. Indeed, the online transcoding needs the CPU while the content is being watched and fills the storage space progressively whereas the offline transcoding implies this processing was already done: the CPU only needs to process the network transactions (and background system operations of course) but the storage is already filled, leaving no room for flexibility if the video demand changes over time (and the video database is not updated accordingly).

This last element can be an argument for pushing the utilization of online transcoding on edge nodes for video streaming.

Nevertheless, our experiments lacked diversity as there were only a handful of experiments and a pair of scenarios - or one scenario with a single factor changed being the processing state of the streamed video. We also would have preferred to be able to measure all 5 parameters we planned on measuring as we would have been able to do a sharper comparison by considering the quality of experience.
Now the dynamic switching between video resolution is fixable, these experiments will be doable in the future.