From linuxhippy at gmail.com Sat Apr 11 07:04:31 2020 From: linuxhippy at gmail.com (Clemens Eisserer) Date: Sat, 11 Apr 2020 09:04:31 +0200 Subject: Looking for Shenandoah v2 / ZGC benchmarks Message-ID: Hi, Out of curiosity I am looking for (non-micro) benchmarks comparing Shenandoah 2 with ZGC. >From what can tell looking at the author's slides, obviously both are able to keep pause times extremely low and in a more or less comparable range, overall throughput suffers a bit, but not as drastically as one would expect. However, as far as I can tell, those SpecJBB runs typically presented in slides are in both cases performed on big machines with cores to spare. So from what I understand, throughput results reflect mostly the barrier code in the application threads and the additional memory bandwith pressure caused by concurrent compaction, however there are additional cores which can absorb at least part of the additional concurrent load introduced by concurrent compaction. What those benchmarks do not seem to reflect however is the total amount of instructions executed, or better, CPU cycles consumed, i.e. cloud costs. Does anybody know of benchmarks / comparisons of ZGC / Shenandoah that offer more than just throughput / pause time measurements? Thanks, Clemens