Github user jerrypeng commented on the issue:
https://github.com/apache/storm/pull/2400
@revans2 interesting work! Though the formula used for the
DefaultSchedulingPriorityStrategy seems to have a bias towards smaller
topologies.
For example:
Total amount of Resource R in cluster= 1000
Topology 1 Resource R: Request = 100 Guarantee = 200
Topology 2 Resource R: Request = 10 Guarantee = 20
Topology 1 score = (100 - 0 - 200) / 1000 = 0.1
Topology 2 score = (10 - 0 - 20) / 1000 = 0.01
Topology 2 will be prioritized or topology one correct? Even though the
ratio their respective request and guarantee is the same. This may cause
larger topologies to be starved.
---
