Check Minerva, RobustMpc & Pensieve experiments

[Alexandru Dan]

Write down in this issue the formats for the results from the papers above. This might help in our final set of experiments for evaluation.

[Alexandru Dan]

Minerva:

• Min-max fairness alone
  Measures:
  - min-max fairness
  - total quality
  - total rebuffering

  • "We sample 48 videos uniformly at random from our corpus to play over a link with a constant total capacity of 8 Mbit/s. Video start times are determined by a Poisson process such that, on average, 4 videos share the link at a given time."

• Fairness with cross traffic

  • "To evaluate Minerva on connection-level fairness, we play multiple Minerva videos, again chosen uniformly at random, simultaneously with the same N videos running over Cubic."

• Fairness with "variable" cross traffic(TCP short flows)

  • "We emulate web traffic from 200 servers, drawing flow sizes froman empirical distribution derived from CAIDA data [1]. The crosstraffic’s average offered load varies between 4 and 16 Mbit/s on a20 Mbit/s emulated link."
  • Note they measure the performance of running a single video over a fixed link while the bandwidth is shared with emulated web traffic from 200 servers(which sums 4-16Mbit/s).

[Alexandru Dan]

RobustMPC and Dynamic basically use traces. Environment is Mahimahi for both Minerva and RobustMPC.