8335347: GenShen: Revert change that has adaptive heuristic ignore abbreviated cycles

[earthling-amzn]

Additional performance testing has shown no benefit from this feature. In fact, latency is degraded on the extremem benchmark with this feature enabled.

---

Progress

• Change must not contain extraneous whitespace
• Commit message must refer to an issue
• Change must be properly reviewed (1 review required, with at least 1 Committer)

Issue

• JDK-8335347: GenShen: Revert change that has adaptive heuristic ignore abbreviated cycles (Bug - P4)

Reviewers

• Kelvin Nilsen (@kdnilsen - Committer)

Reviewing

Using git

Checkout this PR locally:
$ git fetch https://git.openjdk.org/shenandoah.git pull/455/head:pull/455
$ git checkout pull/455

Update a local copy of the PR:
$ git checkout pull/455
$ git pull https://git.openjdk.org/shenandoah.git pull/455/head

Using Skara CLI tools

Checkout this PR locally:
$ git pr checkout 455

View PR using the GUI difftool:
$ git pr show -t 455

Using diff file

Download this PR as a diff file:
https://git.openjdk.org/shenandoah/pull/455.diff

Webrev

Link to Webrev Comment

[bridgekeeper]

👋 Welcome back wkemper! A progress list of the required criteria for merging this PR into master will be added to the body of your pull request. There are additional pull request commands available for use with this pull request.

[openjdk]

@earthling-amzn This change now passes all automated pre-integration checks.

ℹ️ This project also has non-automated pre-integration requirements. Please see the file CONTRIBUTING.md for details.

After integration, the commit message for the final commit will be:

8335347: GenShen: Revert change that has adaptive heuristic ignore abbreviated cycles

Reviewed-by: kdnilsen

You can use pull request commands such as /summary, /contributor and /issue to adjust it as needed.

At the time when this comment was updated there had been no new commits pushed to the master branch. If another commit should be pushed before you perform the /integrate command, your PR will be automatically rebased. If you prefer to avoid any potential automatic rebasing, please check the documentation for the /integrate command for further details.

➡️ To integrate this PR with the above commit message to the master branch, type /integrate in a new comment.

[mlbridge]

Webrevs

• 00: Full (64a84a27)

[kdnilsen]

Thanks.

[earthling-amzn]

/integrate

[openjdk]

Going to push as commit 7542e90.

[openjdk]

@earthling-amzn Pushed as commit 7542e90.

💡 You may see a message that your pull request was closed with unmerged commits. This can be safely ignored.

[ysramakrishna]

Additional performance testing has shown no benefit from this feature. In fact, latency is degraded on the extremem benchmark with this feature enabled.

Did we compare degenerated cycle count for more spiky or variable load benchmarks, where the original fix was supposed to help.

I also wonder if instead of the "bang, bang" control (completely ignore abbreviated as was the previous fix, now withdrawn vs treat abbreviated numbers on par with non-abbreviated cycles), I wonder if you might maintain two decaying averages for the two types separately, then toss a coin whose probability is based on the decaying ratio of the two types of events, and use the decaying average value of the corresponding type or the other based on the result of the coin toss. In effect, you are simulating and trying to predict the probability of a long or short cycle, and using the appropriate measure. Not sure how that would behave since the value seems to be continuously calculated, it would probably effectively revert to the same value as the decaying average gives without doing this more complex simulation. :-)

Anyway, was just thinking out loud here...

[earthling-amzn]

We tried every benchmark in the pipeline (dacapo, specjvm2008, specjbb2015, heapothesys, extremem, dilluvian). I think this attempt at optimization didn't work because an application is going to either have many abbreviated cycles, in which case, the heuristic lives in ignorance and runs based on the time for the rare regular length cycle (making it too aggressive). Or, the application may have very few abbreviated cycles, in which case this optimization doesn't matter. I'm sure we could craft a workload that created a sequence of abbreviated cycles, and then caused a non-abbreviated cycle, but this seems uncommon in practice. In those situations, if a degenerated cycle did occur, the late penalties would make the heuristic more aggressive. Your idea is interesting and perhaps something we could experiment with in the fullness of time.