ampere4 processing 150arm64-quarterly has been building just:

qt6-webengine-6.10.1_2  (Elapsed: 18:52:54)
chromium-145.0.7632.109 (Elapsed: 18:51:49)
electron37-37.10.3      (Elapsed: 18:13:35)

for 3+ hours: after Feb 21 17:10:50 UTC 2026
(now: 20:12:??), as an indication of context for
the below.

Example Load Averages (from from the same window update):
( 6%) 9.55 9.61 9.37

The Swapinfo figure that I've been seeing when I look: 54.10%
(Without a swap space size, I can not convert to RAM+SWAP usage figures,
even with RAM known.)

I wonder what the paging activity scale is. Given the 160 FreeBSD cpus
with under 10 in use, load averages being higher than the CPU count
available do not explain any notable change in build times vs. when,
say, qt6-webengine-6.10.1_2 builds without either of the other 2 also
building at the same time.

(Sometimes mongodb70-armv80a-7.0.28 ends up in parallel.)


143arm64-quarterly (presuming similarity for the swap space usage and
load average contributions per active builder) . . .

qt6-webengine-6.10.1_2: 20:10:22
(no electron* built; chromium started after qt6-webengine finished)

qt6-webengine-6.10.1_2: 26:35:47
(electron37-37.10.3 built in parallel; chromium not built)

(Focusing on the time ratio, not the absolute times.)

It would be interesting if poudriere(-devel) had a (by example):

JUST_ONE_BUILDER_AT_A_TIME="qt6-webengine chromium electron37"

So: each mutually exclusive with everything else. That, along with
having per-such-port MAKE_JOBS_NUMBER_LIMIT (or the like) allowing more
than the official builder default of 3 parallel makes, say via a make.conf .


-- 
===
Mark Millard
marklmi at yahoo.com


Reply via email to