The post misleads readers into thinking that JVM runs the GC before exit. It does not. When I was writing the Epsilon JEP, I meant that it might be futile to have a hundreds-of-ms-long GC cycle, when the program exits…
Oracle is in a separate category, no question there. It was my mistake to mention other vendors in the thread that is about AdoptOpenJDK. I just wanted to show where the AdoptOpenJDK sits in the larger OpenJDK…
Granted, Oracle does contribute lots of code. As the result, lots of Oracle-created code gets backported too. That said, I see people backport a lot of code/tests that they wrote themselves, and/or with the help of…
There are many OpenJDK(-based) distributions out there, mostly built from the OpenJDK {8, 11} Update Releases sources. AdoptOpenJDK provides one of those distributions. Most of the actual backport work is done by the…
Red Hat also provides binaries built from the unmodified OpenJDK source here: https://adoptopenjdk.net/upstream.html -- which could be downloaded without developer account.
In tracing collectors, marking (and generally walking the heap in possibly random order) would take a significant amount of time. After marking is done, you may decide to move only a few objects, so the overhead of the…
I agree with you and present even a stronger statement: talking about GC pauses without measuring the actual end-to-end latencies may give the false impression! The overall throughput might look good, but threads might…
> throttling pauses are not counted as GC pauses Yes. All GCs I know employ the negative feedback loops between allocators and GCs. Stop-the-world GCs are on the far side of that spectrum: there, allocation failure…
The pause time behavior depends a lot on what is going on in application itself. Shenandoah wiki has a section on that: https://wiki.openjdk.java.net/display/shenandoah/Main#Main-G.... It also depends on the hardware,…
Author's note: Interesting to see this resurfacing after a year. JMH had improved significantly since then, including the release of 1.0.
The post misleads readers into thinking that JVM runs the GC before exit. It does not. When I was writing the Epsilon JEP, I meant that it might be futile to have a hundreds-of-ms-long GC cycle, when the program exits…
Oracle is in a separate category, no question there. It was my mistake to mention other vendors in the thread that is about AdoptOpenJDK. I just wanted to show where the AdoptOpenJDK sits in the larger OpenJDK…
Granted, Oracle does contribute lots of code. As the result, lots of Oracle-created code gets backported too. That said, I see people backport a lot of code/tests that they wrote themselves, and/or with the help of…
There are many OpenJDK(-based) distributions out there, mostly built from the OpenJDK {8, 11} Update Releases sources. AdoptOpenJDK provides one of those distributions. Most of the actual backport work is done by the…
Red Hat also provides binaries built from the unmodified OpenJDK source here: https://adoptopenjdk.net/upstream.html -- which could be downloaded without developer account.
In tracing collectors, marking (and generally walking the heap in possibly random order) would take a significant amount of time. After marking is done, you may decide to move only a few objects, so the overhead of the…
I agree with you and present even a stronger statement: talking about GC pauses without measuring the actual end-to-end latencies may give the false impression! The overall throughput might look good, but threads might…
> throttling pauses are not counted as GC pauses Yes. All GCs I know employ the negative feedback loops between allocators and GCs. Stop-the-world GCs are on the far side of that spectrum: there, allocation failure…
The pause time behavior depends a lot on what is going on in application itself. Shenandoah wiki has a section on that: https://wiki.openjdk.java.net/display/shenandoah/Main#Main-G.... It also depends on the hardware,…
Author's note: Interesting to see this resurfacing after a year. JMH had improved significantly since then, including the release of 1.0.