I’m running a two hour benchmark using production log replay right now. The CMS hosts have between 3% and 7% GC overhead (the portion of their CPU time spent in GC).
The G1 host has around 1% GC overhead. I accidentally started one host with the throughput collector. That has 02% overhead and has a clearly faster response time. The other hosts are averaging from 460 to 660 ms. The host with the throughput collector is at 390 ms. The 95th percentile is also faster. And it is using less CPU. Isn’t GC tuning fun? wunder Walter Underwood wun...@wunderwood.org http://observer.wunderwood.org/ (my blog) > On Jan 23, 2017, at 2:49 PM, Pushkar Raste <pushkar.ra...@gmail.com> wrote: > > Hi Walter, > We have been using G1GC for more than a year now and are very happy with > it. > > The only flag we have enabled is 'ParallelRefProcEnabled' > > On Jan 23, 2017 3:00 PM, "Walter Underwood" <wun...@wunderwood.org> wrote: > >> We have a workload with very long queries, and that can drive the CMS >> collector into using about 20% of the CPU time. So I’m ready to try G1 on a >> couple of replicas and see what happens. I’ve already upgraded to Java 8 >> update 121. >> >> I’ve read these pages: >> >> https://wiki.apache.org/solr/ShawnHeisey#G1_.28Garbage_First.29_Collector >> <https://wiki.apache.org/solr/ShawnHeisey#G1_.28Garbage_First.29_Collector >>> >> https://gist.github.com/rockagen/e6d28244e1d540c05144370d6a64ba66 < >> https://gist.github.com/rockagen/e6d28244e1d540c05144370d6a64ba66> >> >> Any updates on recommended settings? >> >> wunder >> Walter Underwood >> wun...@wunderwood.org >> http://observer.wunderwood.org/ (my blog) >> >> >>
