Re: Periodic rant about SCHED_ULE

From: Mark Millard <marklmi_at_yahoo.com>
Date: Thu, 23 Mar 2023 18:48:18 UTC
This goes the direction of comparing SCHED_ULE vs. SCHED_4BSD
for a poudriere bulk build sequence, no competing dnetc or the
like involved. This is on the same systems as prior buildworld
buildkernel and dnetc testing. The style of building allows
large load averages compared to the 32 hardware threads
(16 ThreadRipper 1950X cores). It is a root on ZFS context.

Summary: SCHED_4BSD vs. SCHED_ULE makes little difference
         for this kind of context.



SCHED_4BSD:

# poudriere bulk -jmain-amd64-bulk_a -c -f ~/origins/amd64-origins.txt
. . .
[main-amd64-bulk_a-default] [2023-03-23_00h23m37s] [committing:] Queued: 541 Built: 541 Failed: 0   Skipped: 0   Ignored: 0   Fetched: 0   Tobuild: 0    Time: 03:32:46

So: a little over 3.5 hr.

For maximum observed load averages, MaxObs: 197.37, 159.25, 110.71

It did use a little swap space:

Swap: 491520Mi Total, . . ., 194184Ki MaxObsUsed, 111420Mi MaxObs(Act+Lndry+SwapUsed)
, 119490Mi MaxObs(Act+Wir+Lndry+SwapUsed)



SCHED_ULE:

# poudriere bulk -jmain-amd64-bulk_a -c -f ~/origins/amd64-origins.txt
. . .
[main-amd64-bulk_a-default] [2023-03-23_08h04m39s] [committing:] Queued: 541 Built: 541 Failed: 0   Skipped: 0   Ignored: 0   Fetched: 0   Tobuild: 0    Time: 03:28:00

So: a little under 3.5 hr.
(Somewhat under 5 minutes less than for SCHED_4BSD.)

For maximum observed load averages, MaxObs: 203.63, 156.13, 118.31

It did not use swap space (no MaxObsUsed reported):

Swap: 491520Mi Total, . . ., 90569Mi MaxObs(Act+Lndry+SwapUsed)
, 101048Mi MaxObs(Act+Wir+Lndry+SwapUsed)

===
Mark Millard
marklmi at yahoo.com