ZGC: The Future of Low-Latency Garbage Collection Is Here 

good question but if they are saying 4 times increase that means it is the best GC in terms of throughput. or may be other GCs will get some benefits from this learning.

I think the idea is reusing old objects instead of deallocating and allocating

I don't think reference counting is a good fit for the existing infrastructure

Why would you want to have the the overhead? If you know the lifecycle at compile time, you know when you can free the memory anyway right? Then you don't need the memory overhead to store the reference counts, nor the cpu to update them. Reference counting isn't a silver bullet...

@@cptfwiffo I agree with what you're saying. But may be there was some confusion as to what "reference counting" is. More specifically I meant compile-time reference counting (not runtime reference counting) For example: void foo() { var x = new StringBuilder("Hello"); x.append(", world!"); System.out.println(x.toString()); // Compiler to inject code to free x } Here the compiler needs to count the references in order to ensure that references to the new StringBuilder is limited to local variables before injecting code to free it (it could of course do more complicated analysis). My understanding is that, this is not being done by javac.

@@slr150 In your example x never escapes so the compiler can explode it into its components and no object is allocated at all