[jdk16] RFR: 8257974: Regression 21% in DaCapo-lusearch-large after JDK-8236926 [v2]

[Stefan Johansson]

On Thu, 17 Dec 2020 14:51:39 GMT, Claes Redestad <redestad at openjdk.org> wrote:

> LGTM - seems like a reasonable approach. I'm sure an argument can be made for being able to control the delays (I can picture situations where tuning in either direction might make sense), but I agree with fixing this regression by making these delays constant for now.

Thanks for the review.

Yes, we might end up adding flags for these delays in a later release. Or we might tune the values for the constants if we find something that seem to work better. As long as we don't see this causing big problems, I would prefer if we can have well balanced constants, instead of adding flags that don't add much value.

-------------

PR: https://git.openjdk.java.net/jdk16/pull/42