git: 0a99422970d7 - main - Move mips and arm to 1000Hz by default.

[Warner Losh]

> On Jun 17, 2021, at 3:31 AM, Gary Jennejohn <gljennjohn at gmail.com> wrote:
> 
> On Thu, 17 Jun 2021 09:20:46 +0000
> Alexey Dokuchaev <danfe at freebsd.org> wrote:
> 
>> On Thu, Jun 17, 2021 at 02:01:05AM +0000, Warner Losh wrote:
>>> commit 0a99422970d7fbdb1e17724339a8dc37082f3431
>>> 
>>>    Move mips and arm to 1000Hz by default.
>>> 
>>>    armv6 and armv7 systems already were 1000Hz. The other armv5 were a
>>>    mix of 100 and 1000. This changes them to 1000. Should there be
>>>    issues, we can add options HZ=100 to the systems that have bad
>>>    performance at the drop of a hat.
>> 
>> Do we have this 1000Hz vs. 100Hz thing documented somewhere for those
>> who are thinking about tuning it?  There are various "optimization"
>> howto's floating on the Internet here and there which recommend this
>> change, so it would be nice to have some definitive documentation on
>> the subject, e.g. in our Handbook or Wiki.
>> 
> 
> It's discussed in /sys/conf/NOTES, but most users probably wouldn't
> think about looking there.  An entry in the Handbook or Wiki could
> be useful.

I’ve brought over NetBSD’s hz(9) and hardclock(9) man pages in https://reviews.freebsd.org/D30802 with adjustments for FreeBSD. I mention the kern.hz tunable and the fact that it default to 100 for VM guests. This should cover people’s curiosity as well as plug the hole in our docs. All this stuff was from 4.4BSD, so NetBSD’s man pages were a good place to start.

The HZ kernel option is something almost nobody should be adjusting. I’ve updated the text slightly in NOTES in https://reviews.freebsd.org/D30804 if people want to tweak it further before I commit. Wiki is terrible for this stuff, since it goes stale too quickly, but I’d be happy to review someone else’s changes to the handbook for tuning this value (though honestly, it should almost never be frobbed). kern.hz likely shouldn’t be adjusted either, but it’s a much easier knob to turn than rebuilding the kernel with a new HZ and there’s more cases where it can be useful (though tests I’ve done in the past suggest values > 10,000-50,000 might not be that useful).

Finally, NetBSD has an option(4). We’re long past the time where we should consider it. I’d love to review changes for that too. Most of a first cut on text can be stolen from NOTES, if someone is looking for an easy project (it need not be complete to be useful, at least initially)

Warner
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 874 bytes
Desc: Message signed with OpenPGP
URL: <http://lists.freebsd.org/pipermail/dev-commits-src-all/attachments/20210617/e7fb1973/attachment.sig>