On Thu, 18 May 2023 11:02:21 GMT, Yi Yang <yy...@openjdk.org> wrote: > I'm interested in if it is faster in STW time with the parallel write to use > compression or not? (Does not compressing take time, or does compressing mean > we write fewer bytes, so ends up being faster.)
I updated the benchmark data when turning compression on. In general, this patch significantly reduces 73~80% application pause time. When compression is enabled, STW/Total time heavily depends on the sparseness of the application heap. If the heap is full of compressible(e.g. all objects are empty byte array), Total ≈ STW, the merge process is incredibly fast. If the heap data is not suitable for compression(e.g. all objects are full of random data), the STW reduction is not appealing, the total dump time is also increased. P.S. After further consideration, I think the title "Enhance heap dump" is also not quite accurate. This patch has fundamentally changed the behavior of heap dump, where before it was used to generate a complete dump, now it is used to generate multiple dump files and then combine them into a single dump. I am currently considering a more appropriate title. Thanks. ------------- PR Comment: https://git.openjdk.org/jdk/pull/13667#issuecomment-1556473626
