G1 is Java's default garbage collector in most environments, and its throughput has been considerably improved in JDK 26 by streamlining its write barriers. This conversation explores the background of that change and dives deep into regions, write barriers, concurrent marking, card tables, and how all that impacts throughput before eventually getting to the improvements made in Java 26, which lay further groundwork to G1 becoming the one and only default collector in _all_ environments.

In this "Ask the Architect" episode of the Inside Java Podcast, recorded during JavaOne 2026, Nicolai Parlog talks to Stefan Johansson, HotSpot Garbage Collection engineer at Oracle.

JEP 522: https://openjdk.org/jeps/522

~~ Chapters ~~

0:00 Intro
0:45 Conversation with Stefan
16:40 Outro

Make sure to check https://inside.java/podcast for more episodes and the https://www.youtube.com/playlist?list=PLX8CzqL3ArzUMVSzm-z_-if8BIB55EGl4 playlist.


Tags: #Java #OpenJDK #Performance

Java (Official Oracle)

A deep-dive podcast episode with Stefan Johansson, a HotSpot GC engineer at Oracle, covering G1 garbage collector internals and the throughput improvements coming in JDK 26 via JEP 522. The discussion covers write barriers, card tables, concurrent marking, concurrent refinement, and the synchronization overhead that has historically made G1 slower than Parallel GC. The key improvement in JDK 26 introduces a dual card table design that replaces fine-grained per-write synchronization with coarser-grained handshaking when switching between tables, reducing barrier overhead with only ~0.2% additional native memory cost. The episode also touches on JEP 523, which aims to make G1 the universal default GC across all configurations including small heaps and single-CPU environments.

How JDK 26 Improves G1's Throughput - Inside Java Podcast 54