Gabriel Scherer
Gabriel Scherer
> The idle domain at least needs to scan its roots. We could shortcut this part whenever the minor heap is completely empty. (But there is still other shared work...
Thanks! Be careful in interpreting the GC statistics above that I think the GC-statistics code is currently buggy. In fact many of the stats you show contain the bug that...
I pushed my little benchmark at https://gitlab.com/gasche/idle-domains-bench On my Linux machine with "4 cores but 8 threads" (`recommended_domain_count` is 8), with Firefox and what not running at the same time,...
> But, would the waiting domain's minor heap be entirely empty in this case? IIUC, it may not be the case when something else is going on before the waiting...
Just to be clear, I don't want to insist on having idle domains, and I think that your idea of having the API that maximizes performance, possibly at the cost...
I am thinking that I could post my benchmark sources on Discuss and encourage interested people to run their own experiments. (We would get 50 results of 50 shoddy experiments,...
Here are results of my benchmarks with an implementation of "idle domains" that still does a bit of work; ``` w01: [i00 2.41s (x1.04)] [i01 2.79s (x1.20)] [i02 2.81s (x1.21)]...
@dra27 if/when you have time, checking that 4.07 doesn't exhibit this behavior would be nice :-)
## Current status of unboxed constructors In case it can help, here is all the information one could want on the current status of unboxed constructors: - My work with...
Sounds great, thanks!