Compile time slower in version 0.11.7-86-18d144

[romain-gilles-ultra]

Hi Mill,
I updated my CI with the mill version 0.11.7-86-18d144 and noticed a performance decrease in the compilation time in parallel mode.
I was on version 0.11.7-70-654f58 before that. Here are the numbers:

Before update

Version: 0.11.7-70-654f58

./mill __.compile => 5m 36s
./mill -j 4 __.compile => 2m 40s

After update

Version: 0.11.7-86-18d144

./mill __.compile => 5m 34s
./mill -j 4 __.compile => 3m 28s

Issue

The compile time between 0.11.7-70-654f58 and 0.11.7-86-18d144 increased by 30% on my build in parallel mode

[lefou]

654f58...18d144

[lefou]

> echo -n "0.11.7-70-654f58" > .config/mill-version && mill -i -j 0 __.prepareOffline + clean __.compile && time mill -i -j 4 __.compile
...
real    1m18,300s
user    12m54,230s
sys     0m25,926s

> echo -n "0.11.7-86-18d144" > .config/mill-version && mill -i -j 0 __.prepareOffline + clean __.compile && time mill -i -j 4 __.compile
...
real    1m26,116s
user    13m39,047s
sys     0m24,853s

> echo -n "0.11.7-70-654f58" > .config/mill-version && mill -i -j 0 __.prepareOffline + clean __.compile && time mill -i -j 4 __.compile
...
real    1m13,489s
user    12m7,180s
sys     0m19,427s

> echo -n "0.11.7-86-18d144" > .config/mill-version && mill -i -j 0 __.prepareOffline + clean __.compile && time mill -i -j 4 __.compile
...
real    1m13,100s
user    12m11,088s
sys     0m19,742s

I can't notice any significant difference on my laptop when compiling the full Mill repo. I think the most relevant change is:

• Update zinc from 1.9.6 to 1.10.0, use improved analysis store #2899

Beside some improvements and changes regarding over- and undercompilation, zinc 1.10 introduced a new analysis file store which is itself using multiple threads. I already limited its concurrency to 8 threads (which was chosen arbitrarily). But it could be relevant, given the different OS and physical core count.