Github user jmartell7 commented on the issue:
https://github.com/apache/storm/pull/2399
@HeartSaVioR , I've updated the title. The overhead of serializing the
local tuples to measure the size didn't seem worth it, and local tuples
generally become a smaller fraction of the total as the number of workers
increases.
I'm working on performance numbers using ThroughputVsLatency, but am open
to other ideas. ---
