Allows better handling of streams with wildly variable throughputs, by preventing the auto-adaption from going too high when the throughput is high, which would cause problems if the throughput drops suddenly. The new low throughput would mean we might take an extremely long time to hit the adapted check interval (that is, much longer than the desired/specified targetInterval) and re-adapt and/or display the throughput as desired.
The downside is that when the maxCheckInterval is used (it isn't enabled by default), if the throughput is steadily quite high, the check interval won't be able to auto-adapt any higher than this max, causing many needless TOD checks and potentially increasing tuple latency more than necessary.
Allows better handling of streams with wildly variable throughputs, by preventing the auto-adaption from going too high when the throughput is high, which would cause problems if the throughput drops suddenly. The new low throughput would mean we might take an extremely long time to hit the adapted check interval (that is, much longer than the desired/specified targetInterval) and re-adapt and/or display the throughput as desired.
The downside is that when the maxCheckInterval is used (it isn't enabled by default), if the throughput is steadily quite high, the check interval won't be able to auto-adapt any higher than this max, causing many needless TOD checks and potentially increasing tuple latency more than necessary.