RFR: Take boost mode into account when calculating duration of gc

Per Liden per.liden at oracle.com
Tue Mar 6 15:42:53 UTC 2018


Got an off-line request from Stefan to make it more clear that we're 
dealing with a normalized duration, which I agree is good. So here's an 
updated patch which renames the "duration" value and function to 
"normalized_duration".

Incremental diff: 
http://cr.openjdk.java.net/~pliden/zgc/normalized_duration/webrev.0vs1/

Full diff: 
http://cr.openjdk.java.net/~pliden/zgc/normalized_duration/webrev.1/

/Per

On 03/05/2018 03:24 PM, Per Liden wrote:
> When using boost mode (i.e. when we're using all available worker 
> threads instead of ConcGCThreads) we typically complete a GC cycle much 
> faster than normal (as intended). However, this causes the "duration of 
> gc" calculations to be artificially deflated, can later lead to a 
> situation where we risk starting a GC cycle too late. This patch 
> corrects the "duration of gc" calculation by taking the "boost factor" 
> into account.
> 
> http://cr.openjdk.java.net/~pliden/zgc/normalized_duration/webrev.0/
> 
> /Per


More information about the zgc-dev mailing list