Date: Wed, 9 May 2007 17:30:28 +0100 From: Thomas Hurst <tom.hurst@clara.net> To: Bill LeFebvre <bill@lefebvre.org> Cc: freebsd-stable@freebsd.org Subject: Re: top shows '<swapped>' Message-ID: <20070509163028.GA73424@voi.aagh.net> In-Reply-To: <4641CFDA.9020805@lefebvre.org> References: <d018a9bd0705080607x402b29a2q4d5aae29454626d7@mail.gmail.com> <4641CFDA.9020805@lefebvre.org>
next in thread | previous in thread | raw e-mail | index | archive | help
* Bill LeFebvre (bill@lefebvre.org) wrote:
> The <> are only used when the process flag PS_INMEM is clear, which
> is supposed to indicate that the process is or is not "in memory".
> This flag is only ever cleared in swapout, called from swapout_procs.
> My bet is that the processes are being marked for swap but the dirty
> pages never actually go anywhere since you don't have a backing
> store. Maybe someone more familiar with the inner workings of the VM
> system can fill us in on what happens on a system with no swap.
I'm seeing this sort of thing too -- I do have swap, but it's not being
used by these processes (swapoff -a didn't do anything to them):
Mem: 1672M Active, 5337M Inact, 279M Wired, 400M Cache, 215M Buf, 74M Free
Swap: 10G Total, 12K Used, 10G Free
1251 www 1 4 0 87884K 0K accept 2 0:00 0.00% <httpd>
1106 root 1 20 0 12756K 0K pause 1 0:00 0.00% <smbd>
950 root 1 115 0 8536K 0K select 3 0:00 0.00% <pure-ftpd>
1143 mysql 1 8 0 5220K 0K wait 3 0:00 0.00% <sh>
1288 root 1 5 0 3644K 0K ttyin 2 0:00 0.00% <getty>
The bulk of the data is probably "swapped" to the on-disk binaries, but
this would imply there isn't a single page unique to each process.
Quite why it's bothering in the first place with >5GB Inact I'm not
sure -- is it unmapping idle processes to conserve VM objects?
I also find it interesting that I only noticed this behavior a few days
ago and suddenly someone else mentions it too :)
--
Thomas 'Freaky' Hurst
http://hur.st/
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070509163028.GA73424>