Date: Tue, 2 Jun 1998 19:24:24 +0400 From: "Mikhail A. Sokolov" <mishania@demos.su> To: Garrett Wollman <wollman@khavrinen.lcs.mit.edu> Cc: dg@root.com, freebsd-current@FreeBSD.ORG Subject: Re: mbuf cluster problem continues!! Message-ID: <19980602192424.02082@demos.su> In-Reply-To: <199806011626.MAA22057@khavrinen.lcs.mit.edu>; from Garrett Wollman <wollman@khavrinen.lcs.mit.edu> on Mon, Jun 01, 1998 at 12:26:41PM -0400 References: <015b01bd8cf4$23f4da40$e34a05cb@alpine.iaccess> <199806010520.WAA09567@implode.root.com> <199806011626.MAA22057@khavrinen.lcs.mit.edu>
next in thread | previous in thread | raw e-mail | index | archive | help
Hello,
On Mon, Jun 01, 1998 at 12:26:41PM -0400, Garrett Wollman wrote:
# <<On Sun, 31 May 1998 22:20:10 -0700, David Greenman <dg@root.com> said:
# > I've seen several reports of mbuf leaks in the specific case of running
# > squid proxy servers.
# Not seen here.
# root@xyz(4)$ netstat -m
# 825/1408 mbufs in use:
Oh well, it's not squid what is definite culprit here, not closing tcp
connections: let's take a machine, which is attacked by clients, is being
agressively used nfs server and doesn't even have any services besides nfs,
which could leave tcp connections not closed:
{zz}~/# netstat -m
10577/10688 mbufs in use:
10057 mbufs allocated to data
520 mbufs allocated to packet headers
451/728 mbuf clusters in use
2792 Kbytes allocated to network (79% in use)
0 requests for memory denied
0 requests for memory delayed
0 calls to protocol drain routines
{zz}~/# uptime
6:44ÐÐ up 1 day, 8:57, 21 users, load averages: 1.85, 1.80, 1.62
NMBCLUSTERS are 24k, and still the machine will leak mbuf's once a week.
{zz}~/# nfsstat -w 1
Getattr Lookup Readlink Read Write Rename Access Readdir
Client: 0 0 0 0 0 0 0 0
Server: 4919834 3679844 37708 1306501 1102222 17918 121153149 48113
...
This machine usually runs some 20-200 simultaneous sendmails, 20-100 interactive
(read: ~shell) users, up to 5 nfs clients, ~50 simulateneous httpd's.
Taking a look at it's neighbour, a virtual web/ftp server:
{xx}~/# netstat -m
1813/4256 mbufs in use:
1203 mbufs allocated to data
609 mbufs allocated to packet headers
1 mbufs allocated to socket names and addresses
1150/2802 mbuf clusters in use
6136 Kbytes allocated to network (41% in use)
0 requests for memory denied
0 requests for memory delayed
0 calls to protocol drain routines
{xx}~/# uptime
7:07ÐÐ up 40 days, 3:11, 11 users, load averages: 1.07, 0.89, 0.96
{xx}~/# nfsstat -w 1
Getattr Lookup Readlink Read Write Rename Access Readdir
Client: 70935468 45165217 772137 7905938 20548126 358237 -2070886142 218656
^^^^^^^^^^^ Ouch.
Interesting, eh? This one is the above mentioned zz nfs client, and the other
difference is that it doesn't handle pop clients and has less sendmail's. Plus,
which is important, it _is_ a caching server (read: squid), serves not that
much, though, some 40k requests/day. Although, it does exceed mbuf's once
in a while, not once per week, though, as you might see from uptime.
{zz}~/# netstat -an
Active Internet connections (including servers)
Proto Recv-Q Send-Q Local Address Foreign Address (state)
tcp 0 0 180.178.31.240.21491 128.50.150.242.64 -225392640
tcp 0 0 148.229.162.242.17651 128.121.206.242.206 -225392640
tcp 0 0 20.85.153.242.18419 128.179.223.242.19 -225392640
^^^^^^^^^^
Charming..
Btw, what's that?
# Of course, it's not really working that hard.
# --
#Garrett A. Wollman | O Siem / We are all family / O Siem / We're all the same
Of course, both are current, as of 1998/04/20. And yes, that's a machines in
productions use; yes, we know we shouldn't... but -stable lacks SMP and
tons of other usefull features.
My 2 kopeks.
--
-mishania
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19980602192424.02082>