shuttie commented on issue #10529: [FLINK-15171] [serialization] fix performance regression caused by too many buffer allocations on string serialization URL: https://github.com/apache/flink/pull/10529#issuecomment-565962443 @pnowojski thanks for the feedback. Current status: * reading the string via thread-local fixed-size buffer is really cool idea, I will make an impl of this today, it should be quite trivial. * most of the char[] allocations come from the readString method: we are allocating a temp buffer to decode characters to, and then the String constructor makes a deep copy of this buffer. Actually there is no need to allocate this buffer on every invocation, we can do the same trick with buffer reuse. The most concerning thing for me is being unable to reproduce the performance regressions on flink-benchmarks. I will try to run them on similar 4-core EC2 instances like c4, hopefully it will help me to understand what's going on. But my main suspect is GC pressure, and both of the planned changes should reduce it quite significantly.
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
