Index | Thread | Search

From:
Mark Patruck <mark@wrapped.cx>
Subject:
Re: kernel rwlocks vs the scheduler
To:
David Gwynne <david@gwynne.id.au>
Cc:
tech@openbsd.org
Date:
Tue, 19 Nov 2024 09:13:43 +0100

Download raw body.

Thread
I'm also running this for a few days now on bare metal and Proxmox VMs
(amd64) and didn't spot any issues.

According to my tests, builds run faster, but mainly because you can
drop in more cores than before that are actually getting used.

For my VMs building kernel, base, x and ports i already did a lot of
testing a few weeks ago after restructuring. I always had best results
with build VMs running 12C, not less but also not more. With 14C i
didn't see any real difference, but >14 build times went up, f.e.
ungoogled-chromium took more like 6h with 32C.

Besides "ungoogled-chromium" which i've built only 3x because of the
more time intense build, everything else went through a few times for
having a better comparison.

Hint: after seeing no real difference wrt build times with 12C + the
rwlocks diff, i directly went up to 16C and continued this way.

Results:
---------

#### before with 12C VM (-j12)
- kernel   = 1m15s - 1m21s
- base     = 40m - 44m
- xenocara = 18m - 21m

- port: ungoogled-chromium = 4h58m - 5h20m


#### with rwlocks diff and 16C VM (-j16)
- kernel   = 1m09s - 1m14s
- base     = 37m - 39m
- xenocara = 16m - 17m

- port: ungoogled-chromium = 4h19m - 4h31m



Especially for big ports it's really an improvement in my environment.
So...instead of buying new hardware, i call you next time ;)

	-Mark

--
Mark Patruck ( mark at wrapped.cx )
GPG key 0xF2865E51 / 187F F6D3 EE04 1DCE 1C74  F644 0D3C F66F F286 5E51
  
https://www.wrapped.cx