Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 7 Nov 2015 22:26:06 +1100 (EST)
From:      Bruce Evans <brde@optusnet.com.au>
To:        Ian Smith <smithi@nimnet.asn.au>
Cc:        Rasool Al-Saadi <ralsaadi@swin.edu.au>,  Hans Petter Selasky <hps@selasky.org>, Luigi Rizzo <rizzo@iet.unipi.it>,  "freebsd-net@freebsd.org" <freebsd-net@freebsd.org>,  Bruce Evans <brde@optusnet.com.au>
Subject:   RE: Timing issue with Dummynet on high kernel timer interrupt
Message-ID:  <20151107220450.X1692@besplex.bde.org>
In-Reply-To: <20151107155144.I12989@sola.nimnet.asn.au>
References:  <6545444AE21C2749939E637E56594CEA3C0DCCC4@gsp-ex02.ds.swin.edu.au> <5638B7B5.3030802@selasky.org> <6545444AE21C2749939E637E56594CEA3C0DE7FF@gsp-ex02.ds.swin.edu.au> <563B2703.5080402@selasky.org> <6545444AE21C2749939E637E56594CEA3C0E0BD9@gsp-ex02.ds.swin.edu.au> <563C6864.2090907@selasky.org> <CA%2BhQ2%2Bhm2z0MkB-8w5xJM7%2Biz13r_ZjwmpZBnb30_D_48gaf-w@mail.gmail.com> <563C786C.1050305@selasky.org> <CA%2BhQ2%2Bj0WiGgzV119M1ZQiXP5Z7fq7deVxDPkOhvTc7hpTETKw@mail.gmail.com> <563CC186.9000807@selasky.org> <6545444AE21C2749939E637E56594CEA3C0E1B79@gsp-ex02.ds.swin.edu.au> <20151107155144.I12989@sola.nimnet.asn.au>

next in thread | previous in thread | raw e-mail | index | archive | help
On Sat, 7 Nov 2015, Ian Smith wrote:

> ...
> I also notice that with HPET, the timer interrupts themselves, whether
> as few as 60-70/s idle in one-shot mode or ~2000/s in periodic, appear
> in systat -vm or vmstat -w10 'Int' column, whereas with LAPIC or (yes I
> even tried) i8254, these timer interrupts don't count in that column.

Interrupt reporting is mostly broken for "fast" interrupts.  E.g.,
systat -v counts them in the Interrupts column (separate and Total) but
not in the Int summary; vmstat -m doesn't count them anywhere (it shows
the Int summary).  vmstat -i shows them much like the Interrupts column
in systat -v (except it actually displays them all when there are too
man CPUs or Interrupts to fit in systat -v).

I use COUNT_IPIS and COUNT_XINVLTLB.  The interrupts counted by these
are actually fast.  These mess up the systat -v display by giving too
many active interrupts to display for just a couple of CPU.  systat
-v tends to display these in preference to more interesting interrupts.
vmstat -i displays these too.

"fast" and fast interrupts should be counted separately and are probaly
best left out of the Int summary.  They should also be displayed separately.

> I'll try Hans' suggestion of adjusting in loader.conf, rebooting between
> various options if really needed, though that'll be far more tedious ..
> [ I've just now seen Bruce's latest, so maybe I haven't been doing the
> wrong thing playing with these after all? ]

The eventtimer choice is readonly, so rebooting seems to be required to
change it.  Too much like Windows.  I often miss being able to change hz.
Event timers now do much more complicated changes than that.  They
implement a sort of dynamic hz, but still use the boot-time hz as a
default.

Bruce



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20151107220450.X1692>