this GC is amazing, I tried it on my low latency app and it's amazing, leave alone the fact the defaults seems a lot better since this GC tends to release more resources to the OS more often.
16:40 The latency is really really impressive. For a more thorough comparison to better understand the tradeoffs I would have loved to see additional graphs showing things like total pause time. Also I wonder how much time is spend waiting for load barriers. It is not a global pause, but also a kind of pause. ZGC is a wonderful addition and its great to see its strengths but I think addressing its weaknesses is just as important. A no pause GC using atomics, locks and metadata would sound nice on paper, but would obviously be uselesss.
When it's ready 🙂 Ultimately I could only provide speculation, that could soon prove incorrect. The most current information would be found on the ZGC dev mailing list: mail.openjdk.org/pipermail/zgc-dev
@@dumdumdumdum8804 JEP for generational ZGC has just been moved to draft status openjdk.org/jeps/439 So that’s a good indication it will be in JDK 21, no promises though 🙂
@@billykorando Thats pretty cool it is already in candidate status. I wonder what will happen to our microservices will it also reduce memory footprint to 1/4 as in the case of Cassandra. We use grpc and kafka most of our services consume 300 MB to 500 MB of memory, with generational zgc i expect it to be 75 to 125MB which will allow us to deploy more services or increase the number of pods for the same service. generational ZGC + lilliput + Leyden -> Huge cost saving.
@@billykorando For Valhalla I think we have to do code changes to take advantage of dense memory layouts, may be if valhalla is retrofitted into autoboxing then probably we will enjoy good performance out of the box by just switching to latest version of jdk.
totally, though on paper zgc should be better than shenan 1. it uses totally free bits of available memory in references, while shenon uses entire extra word per object (though i might be wrong seems from 2.0 it doesnt anymore) 2. it uses only load barriers vs shenon using both load and store gc barriers... Either way we only need to care about 2 GCs to choose from now on... and they are both free.
Java, the language that never stops improving.
This is the JVM not java though
this GC is amazing, I tried it on my low latency app and it's amazing, leave alone the fact the defaults seems a lot better since this GC tends to release more resources to the OS more often.
16:40 The latency is really really impressive. For a more thorough comparison to better understand the tradeoffs I would have loved to see additional graphs showing things like total pause time. Also I wonder how much time is spend waiting for load barriers. It is not a global pause, but also a kind of pause.
ZGC is a wonderful addition and its great to see its strengths but I think addressing its weaknesses is just as important. A no pause GC using atomics, locks and metadata would sound nice on paper, but would obviously be uselesss.
👌💚💚💚💚💚
Very informative presentation. Thanks 👍!
Thank you!
When will generational zgc reach GA.
When it's ready 🙂
Ultimately I could only provide speculation, that could soon prove incorrect. The most current information would be found on the ZGC dev mailing list: mail.openjdk.org/pipermail/zgc-dev
@@billykorando Your Speculation will help, I am totally clueless about it and there isn't much about it in the mailing list of its readiness for GA.
@@dumdumdumdum8804 JEP for generational ZGC has just been moved to draft status openjdk.org/jeps/439
So that’s a good indication it will be in JDK 21, no promises though 🙂
@@billykorando Thats pretty cool it is already in candidate status. I wonder what will happen to our microservices will it also reduce memory footprint to 1/4 as in the case of Cassandra. We use grpc and kafka most of our services consume 300 MB to 500 MB of memory, with generational zgc i expect it to be 75 to 125MB which will allow us to deploy more services or increase the number of pods for the same service.
generational ZGC + lilliput + Leyden -> Huge cost saving.
@@billykorando For Valhalla I think we have to do code changes to take advantage of dense memory layouts, may be if valhalla is retrofitted into autoboxing then probably we will enjoy good performance out of the box by just switching to latest version of jdk.
does this have anything to do with project Leyden?
I think you don't want a tesseract. You want a k-simplex. 🤓
Still in Shenadoah denial? ,)
totally, though on paper zgc should be better than shenan
1. it uses totally free bits of available memory in references, while shenon uses entire extra word per object (though i might be wrong seems from 2.0 it doesnt anymore)
2. it uses only load barriers vs shenon using both load and store gc barriers...
Either way we only need to care about 2 GCs to choose from now on... and they are both free.
Perez George Thompson Steven Rodriguez Anna