(Cross-posted from travis-ci/travis-ci#2999)

Running tests locally to explore this myself...

JRuby 1.7.16 uses upwards of 500MB with default command line where JRuby 9000 used around 750MB but still did complete on my system (Java 8u40). When I gave them each 2GB max heap size, 1.7.16 went as high as 1.05GB and 9k went to 1.28GB, both holding steady once they reached their peaks.

I think it's safe to say two things:

1. For this suite (travis-core) you probably want to bump up the maximum heap to at least 1GB. I'm not sure what all is being retained, but it's a lot in both cases.
2. Based on the base and peak numbers, it looks like 9k is in both cases retaining about 250MB more data for some purpose. I won't speculate until I look at a heap dump.

FWIW, on my system 9k was able to run travis-core's suite in 121s (versus 1.7's 100s) and has 5 failures. So, we've got some good opportunities here and we're already pretty close to 1.7.

I have filed #2270 to track a big memory-retention suspect. This bug will remain open to track the overall heap size difference from 1.7 until we're satisfied we've tidied up.

If I had to speculate we are retaining AST and IR data in 9k. So AST probably is getting captured somewhere and ending up pinned.

no longer relevant ... just like #2270 right?