Date: Mon, 26 Jun 2000 18:22:31 -0600 From: Duane Wessels <wessels@ircache.net> To: freebsd-fs@freebsd.org Subject: vnode/inode starvation Message-ID: <Pine.SGI.4.10.10006261802260.9671-100000@surf.ircache.net>
next in thread | raw e-mail | index | archive | help
Hi,
I have an application -- a caching proxy called Squid. When benchmarking
Squid on FreeBSD-3.5, I'm seeing very high vnode (inode?) usage. After running
for a few hours under peak load, all of the kernel virtual memory (?)
is consumed by vnodes:
Memory statistics by type Type Kern
Type InUse MemUse HighUse Limit Requests Limit Limit Size(s)
256 file desc, devbuf, temp, subproc, vnodes, ifaddr, routetbl, FFS node
FFS node343147 85787K 85787K 85787K 742322 1 0 256
At this point, any process that wants a new vnode is blocked. Top
shows many processes in 'ffsvgt' state, and one process in 'FFS no'
state.
I'm using the RELENG_3 branch, sucked down on June 20:
FreeBSD mr-garrison.measurement-factory.com 3.5-STABLE FreeBSD 3.5-STABLE #4: Tue Jun 20 14:15:04 MDT 2000
root@mr-garrison.measurement-factory.com:/usr/src/sys/compile/SQUID i386
I don't know a whole lot about kernel code and these sorts of low-level
filesystem details. But it seems to me this is really strange. I don't
unerstand why vnodes aren't being reused or freed.
Does vnode reclaimation sort of rely on processes not living very
long? In my case I have six squid processes that never die, each of
which can touch 500,000 or more disk files.
Ideas, references, explanations welcome...
Duane W.
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-fs" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SGI.4.10.10006261802260.9671-100000>