Bug ID: JDK-8076584 G1: Parallelism used for redirty logged cards needs better control

Versions (Unresolved/Resolved/Fixed)

The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.

Other
tbdResolved

The attached plot shows the redirty logged cards time for the standard JDK9 (b45), same with the SemaphoreWait improvements being implemented for lower G1 pauses and SemaphoreWait improvements and redirty logged cards done serially.  The SemaphoreWait work improves the results greatly but doing the work serially improves the pause an additional, significant amount.

The description suggests doing the operation serially was significantly better than alternatives. That may have been a consequence of JDK-8162929. But I agree we should look at the number of threads being used. The cost per card is pretty low, so there have to be a lot of cards per thread to make it worth spinning up more threads. This is similar to what we found with parallel reference processing.
30-08-2019
This is about selecting the correct number of threads or improving thread dispatch for the redirty card operation. JDK-8162929 is about how we store the cards collected during GC.
30-08-2019
I wonder if this was just another version of JDK-8162929? Maybe this is no longer an issue?
30-08-2019
Redirty with 1 GC thread for parallel workers (no semaphoreWait)
09-04-2015
default (no semaphoreWait)
09-04-2015
log for serial redirty (no semaphoreWait)
09-04-2015
log for redirty with MT=1
09-04-2015
Comparisons without the semaphoreWait changes: default, MT=1 for parallel threads used for redirty, serial redirtying.
09-04-2015
redirty logged cards with 1 thread (ParallelGCThreads is the default value but the number of threads used for redirty logged cards was set to 1). The difference between 1 parallel thread and serial code is minimal (indistiguishable in the plots).
02-04-2015
SemaphoreWait and serial redirty logged cards
02-04-2015
SemaphoreWait improvements
02-04-2015
Baseline jdk9 b45
02-04-2015
Redirty logged cards
02-04-2015

Duplicate :	JDK-8327042 - G1: Parallelism used for redirty logged cards needs better control.
Relates :	JDK-8162929 - Enqueuing dirty cards into a single DCQS during GC does not scale
Relates :	JDK-8078117 - Refactor enqueue_discovered_references() and redirty_logged_cards() to share the same work gang.
Relates :	JDK-8076462 - Preserving the referents of concurrent mark work distribution method causes long pause times
Relates :	JDK-8327042 - G1: Parallelism used for redirty logged cards needs better control.