08:43 <kares[m]> sharing class loader for jits based on .rb source file seems to have an impact event when I just run the test:jruby suite - meta-space use is the same but fragmentation is less thus we end up with smaller allocated space 115M vs 95M (class-loader count is down to 10% around 3000 vs 300)

08:45 <kares[m]> will hopefully be able to test this on prod soon (also with jit.max and jit.time.delta)

15:35 <headius[m]> kares: That sounds promising

15:35 <headius[m]> enebo was considering another option where we still use separate classloaders but in groups of ten methods

15:35 <headius[m]> the idea being that we pack as much into that chunk of metaspace as possible but keep the chunks smaller so methods that JIT earlier might age out together

18:04 <enebo[m]> I have a branch I will push tomorrow

18:04 <enebo[m]> 10 per CL did not make the difference I thought but I can get radical reductions using larger values.

18:05 <headius[m]> RADICAL

18:05 <headius[m]> how radical is radical

18:06 <headius[m]> I have additional minor tweaks coming for the indy-literals branch but it could land any time

18:06 <headius[m]> I don't think we have seen anything get slower so far

18:07 <headius[m]> and many things are faster

18:19 <headius[m]> enebo: don't leave me hangin

22:02 <lopex> Numbers

22:02 <lopex> .

22:23 <headius[m]> !!!