You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently all flows are equal, and only the activation time and order of arrival define the order of tasks scheduling.
Prototype a smarter algorithm that provides more fairness:
the activation time for flows should be rounded to the reasonable period (i.e. 1 minute), so that a minor difference like 1s won't be breaking the scoring functions' logic
time wheel's priority queue could use a more complex key, that represents both rounded activation time + a priority score, so that flows are properly grouped by rounded activation time, but stay sorted by score within the same activation time
implement demo of score computing logic:
rarely executed flow goes first (compare schedule periods)
flow with higher derived fanout goes first
In future, the scoring function will be more sophisticated, and will include preference of premium datasets or some sort of QoS guarantee (weight of paid datasets, delay penalty, fairness improvements, average work amount stats)
The text was updated successfully, but these errors were encountered:
Currently all flows are equal, and only the activation time and order of arrival define the order of tasks scheduling.
Prototype a smarter algorithm that provides more fairness:
In future, the scoring function will be more sophisticated, and will include preference of premium datasets or some sort of QoS guarantee (weight of paid datasets, delay penalty, fairness improvements, average work amount stats)
The text was updated successfully, but these errors were encountered: