From: John Kacur <jkacur@redhat.com>
To: Daniel Wagner <dwagner@suse.de>
Cc: Clark Williams <williams@redhat.com>, linux-rt-users@vger.kernel.org
Subject: Re: [rt-tests v1 4/6] cyclictest: Mimik --smp behavior with --affinity
Date: Tue, 26 Jan 2021 11:32:25 -0500 (EST) [thread overview]
Message-ID: <c834b380-8b7b-9bac-4221-e95d69a0dfd0@redhat.com> (raw)
In-Reply-To: <20210126083722.tpyy4jsv25sqzsnk@beryllium.lan>
On Tue, 26 Jan 2021, Daniel Wagner wrote:
> On Tue, Jan 26, 2021 at 12:55:33AM -0500, John Kacur wrote:
> > Well, --smp historically combined -a -t and threads at the same priority
> > You could argue that it is reasonable for -a to automatically imply -t
> > but I have had debates with people about this and we settled on -a just
> > specifies the affinity, and the default number of threads is one unless
> > you use -t.
>
> My thinking is if you set an affinity mask, you want also to assign a
> thread to the CPUs. If not what's the point to set an affinity mask
> without a thread on it. Though the user is able to overwrite this
> by providing the -t options along side the -a option.
Not necessarily, but you want to limit where the threads, no matter
how few or how many run. As I said, it's not an unreasonable default,
but it is not the current default of 1 unless you specify -t
I suggest we stick with the current default in this clean-up effort.
We can discuss if there are some better defaults after this effort is
complete.
>
> > I'm not sure what you mean by this makes cyclictest behave the way the
> > rest of rt-tests does, the rest of rt-tests should match what cyclitest
> > does.
>
> As I wrote, make cyclictest behave the same way as the to other tools
> when -a/-t is used.
>
> > That said, I did some quick runs of signaltest and -a seems broken,
> > sigh.
>
> Is it broken without my patches or with my patches? What is broken?
>
>
I haven't done a git bisect or something of that sort to determine if your
patches broke it, but here's what I see. (running on non-rt laptop now, so
ignore numbers), but it looks like at least the -t option requiring an
argument existed before your changes, but your changes to affinity
probably could use some more testing.
The following is okay
./signaltest -a1-4
2.00 1.52 1.14 2/1079 6541
T: 0 ( 6537) P: 0 C: 14896 Min: 3 Act: 4 Avg: 26 Max: 366
The following is broken, the default without an argument should be 2
./signaltest -t
./signaltest: option requires an argument -- 't'
In the following note that threads other than thread 0 didn't appear
on the screen until after the ctrl-c
./signaltest -t9 -a1-4
1.11 1.14 1.14 2/1075 6864
T: 0 ( 6856) P: 0 C: 1649 Min: 38 Act: 511 Avg: 457 Max: 4911
^CT: 0 ( 6856) P: 0 C: 1660 Min: 38 Act: 554 Avg: 457 Max: 4911
T: 1 ( 6857) P: 0 C: 1660 Min: 38 Act: 490 Avg: 1090 Max: 15024
T: 2 ( 6858) P: 0 C: 1660 Min: 38 Act: 449 Avg: 1091 Max: 15053
T: 3 ( 6859) P: 0 C: 1660 Min: 38 Act: 412 Avg: 1091 Max: 15075
T: 4 ( 6860) P: 0 C: 1660 Min: 38 Act: 378 Avg: 1090 Max: 15081
T: 5 ( 6861) P: 0 C: 1659 Min: 38 Act: 517 Avg: 1091 Max: 15072
T: 6 ( 6862) P: 0 C: 1659 Min: 38 Act: 485 Avg: 1091 Max: 15067
T: 7 ( 6863) P: 0 C: 1659 Min: 38 Act: 522 Avg: 1091 Max: 12832
T: 8 ( 6864) P: 0 C: 1659 Min: 38 Act: 549 Avg: 1091 Max: 12839
etc, it's broken in many ways.
next prev parent reply other threads:[~2021-01-26 18:41 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-12-18 14:19 [rt-tests v1 0/6] libnuma cleanups for cyclictest Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 1/6] cyclictest: Always use libnuma Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 2/6] cyclictest: Use numa API directly Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 3/6] cyclictest: Use affinity_mask for stearing thread placement Daniel Wagner
2021-01-26 5:39 ` John Kacur
2021-01-26 8:41 ` Daniel Wagner
2021-01-26 16:33 ` John Kacur
2021-01-26 17:13 ` Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 4/6] cyclictest: Mimik --smp behavior with --affinity Daniel Wagner
2021-01-26 5:55 ` John Kacur
2021-01-26 8:37 ` Daniel Wagner
2021-01-26 16:32 ` John Kacur [this message]
2021-01-26 17:13 ` Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 5/6] cyclictest: Simplify --smp vs --affinity vs --threads argument logic Daniel Wagner
2020-12-18 14:19 ` [rt-tests v1 6/6] cyclictest: Move verbose message into main Daniel Wagner
2020-12-18 14:41 ` [rt-tests v1 0/6] libnuma cleanups for cyclictest Daniel Wagner
2020-12-18 16:02 ` John Kacur
2020-12-18 16:43 ` Daniel Wagner
2020-12-18 15:57 ` John Kacur
2020-12-18 16:41 ` Daniel Wagner
2020-12-22 17:26 ` Alison Chaiken
2020-12-22 18:04 ` John Kacur
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=c834b380-8b7b-9bac-4221-e95d69a0dfd0@redhat.com \
--to=jkacur@redhat.com \
--cc=dwagner@suse.de \
--cc=linux-rt-users@vger.kernel.org \
--cc=williams@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).