DirectByteBuffers are still using old jdk.internal.ref.Cleaner implementation. That implementation carries a doubly-linked list, and so makes DBB suffer from the same issue fixed for generic java.lang.ref.Cleaner users with JDK-8343704. I think we can migrate DBBs to use java.lang.ref.Cleaner.

With the attached reproducer, we have the similar improvements on DBB churn tests:

% build/macosx-aarch64-server-release/images/jdk/bin/java -Xmx2g -Xlog:gc DBBGC.java

# Before
[16.575s][info][gc] GC(100) Pause Young (Normal) (G1 Evacuation Pause) 1230M->14M(2048M) 48.596ms
[16.756s][info][gc] GC(101) Pause Young (Normal) (G1 Evacuation Pause) 1231M->14M(2048M) 49.211ms
[16.932s][info][gc] GC(102) Pause Young (Normal) (G1 Evacuation Pause) 1231M->14M(2048M) 47.930ms
[17.108s][info][gc] GC(103) Pause Young (Normal) (G1 Evacuation Pause) 1231M->14M(2048M) 47.753ms
[17.292s][info][gc] GC(104) Pause Young (Normal) (G1 Evacuation Pause) 1231M->14M(2048M) 49.631ms
[17.467s][info][gc] GC(105) Pause Young (Normal) (G1 Evacuation Pause) 1231M->14M(2048M) 46.266ms

# After
[6.793s][info][gc] GC(100) Pause Young (Normal) (G1 Evacuation Pause) 595M->9M(989M) 3.811ms
[6.862s][info][gc] GC(101) Pause Young (Normal) (G1 Evacuation Pause) 595M->9M(989M) 3.982ms
[6.930s][info][gc] GC(102) Pause Young (Normal) (G1 Evacuation Pause) 596M->9M(989M) 3.889ms
[6.995s][info][gc] GC(103) Pause Young (Normal) (G1 Evacuation Pause) 596M->9M(989M) 3.810ms
[7.062s][info][gc] GC(104) Pause Young (Normal) (G1 Evacuation Pause) 596M->9M(989M) 3.613ms
[7.126s][info][gc] GC(105) Pause Young (Normal) (G1 Evacuation Pause) 596M->9M(989M) 3.770ms
[7.192s][info][gc] GC(106) Pause Young (Normal) (G1 Evacuation Pause) 596M->9M(989M) 4.012ms

The test is both faster, and takes less GC time.