Date: Sun, 6 Jul 2014 20:55:25 +0300 From: Stefan Parvu <sparvu@systemdatarecorder.org> To: Konstantin Belousov <kostikbel@gmail.com> Cc: freebsd-hackers@freebsd.org Subject: Re: run-queue length question Message-ID: <20140706205525.2b17295699db832fe198de8c@systemdatarecorder.org> In-Reply-To: <20140706123359.GD93733@kib.kiev.ua> References: <20140706143212.3d22d0adfa5dece52de203a3@systemdatarecorder.org> <20140706123359.GD93733@kib.kiev.ua>
next in thread | previous in thread | raw e-mail | index | archive | help
> It is the Linux which follows the traditional definition of the load,
> by counting running, ready to run and 'blocked on the fast i/o' processes
> as adding to the load average. Also, I remember that FreeBSD up to 4.x
> followed this definition.
If I understood right, the classic example follows Solaris SunOS 2.x until Solaris 10
where rq does not include any iowait processes. Hmmm... for sure Solaris 7,8,9
were talking rq with no iowait. I dont remember now about Solaris 10 but I think it does not
add iowait to the rq. And if I recall right Linux 2.4 had similar behaviour as Solaris.
I dont recall FreeBSD but I recall sometime back there was no load average metric
into kernel. I need to read src code for it.
> Sometime during the 5.x rewrites the load was redefined to only count
> running + ready to run threads, which is the current definition, used by LA.
I see. Well Im happy today at least to see FreeBSD having a proper definition
for rq comparative to Linux. Gunther [1] explains in details why rq does include
running and runnable only, very well.
Thanks a lot
[1] - Analyzing Computer System Performance with Perl::PDQ 2nd edition
http://www.perfdynamics.com/books.html
--
Stefan Parvu <sparvu@systemdatarecorder.org>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20140706205525.2b17295699db832fe198de8c>