Not well known is that Heron's performance claims in the "flying faster" blog and SIGMOD paper were made by comparing against a fork of an older (pre apache) version of Storm even though much more performant version was already out.

IMO (biased?) Heron makes surprisingly poor design choices for their key distinguishing features like threading model and backpressure. And it shows when you benchmark it. More specifics: https://github.com/apache/storm/pull/2241#issuecomment-31787...