Re: [Now, other problem] Re: high load cpu with trunk+vlan+carp

tico Mon, 12 Oct 2009 22:41:34 -0700

Luiz Gustavo wrote:

Luiz Gustavo S. Costa wrote:

exists other maillist for the discussion about this ?


dev ? pf ?

please, anybody help-me :)

2009/9/28 Luiz Gustavo S. Costa <luizgust...@mundounix.com.br>:

has anyone with the same problem? developers? what could this be?

2009/9/26 Luiz Gustavo S. Costa <luizgust...@mundounix.com.br>:

Hi people,

Look the interrupt...

# top
load averages:  2.08,  2.62,  1.93
                         19:37:09
23 processes:  2 running, 19 idle, 2 on processor
CPU0 states:  1.0% user,  0.0% nice,  3.8% system, 95.2% interrupt,  0.0% idle
CPU1 states:  4.3% user,  0.0% nice, 41.3% system, 30.4% interrupt, 23.9% idle
Memory: Real: 15M/165M act/tot  Free: 832M  Swap: 0K/5120M used/tot

Very rules of the "rdr", this interrupt stay on high load

Normal operation:
# pfctl -sn | grep rdr | wc -l
     10

With apply the my custom "rdr":
# pfctl -sn | grep rdr | wc -l
    672

basically, i made one rdr for each ip, with below:
rdr pass on vlan30 proto tcp from " . $cliente['ip'] . " to any port
21 tag ftp_" . $nome . " -> 127.0.0.1 port 8021

the "tag" parameter is for queue control of ftp connection

on the freebsd box, this rules works perfectly

any solution ?

2009/9/25 Luiz Gustavo S. Costa <luizgust...@mundounix.com.br>:

Hi all,

Ok, problem resolved !!!

Rules with very "log" configuration, the interface pflog was generate
this high load on interrupt

Now, server in production:
# uptime
12:23PM  up 11 days,  2:16, 3 users, load averages: 0.11, 0.23, 0.48

#top
load averages:  0.13,  0.23,  0.47
                               12:24:04
22 processes:  21 idle, 1 on processor
CPU0 states:  0.0% user,  0.0% nice,  0.0% system, 36.9% interrupt, 63.1% idle
CPU1 states:  0.0% user,  0.0% nice,  0.0% system,  0.0% interrupt,  100% idle

# uname -mprsv
OpenBSD 4.5 GENERIC.MP#0 i386 Intel(R) Pentium(R) 4 CPU 3.00GHz
("GenuineIntel" 686-class)

Migration from one Freebsd

- with Trunk + Vlan + Carp
- Altq (hfsc hack on HFSC_MAX_CLASSES) over Vlan

nice firewall !!!!

2009/9/17 Luiz Gustavo S. Costa <luizgust...@mundounix.com.br>:

Hi guys !

I have one installation of OpenBSD 4.5 with this configuration (network):

(em0 and em1) > trunk0 > vlan[10,30,40,63,65] > carp[10,30,40,63,65]

Ok, this configuration is running perfect !

But,
With traffic on the scenario the cpu is go for down .... i have one
pentium 4 with 2 core, 1 core with 2% idle and 88% of interrupt and
the other core with 40% of interrupt.

ALTQ performance is poor because CPU is high load

I migration this scenario of one freebsd (with lagg and vlan, not
carp) with load average good (no used device polling)

my configs:

[r...@fw2 /usr/src]# sysctl kern.version
kern.version=OpenBSD 4.5-stable (GENERIC.MP) #0: Fri Sep 11 15:34:39 BRT 2009
   r...@fw2.xxxxx.net:/usr/src/sys/arch/i386/compile/GENERIC.MP

[r...@fw2 /usr/src]# pcidump | grep 'Intel PRO'
 6:2:0: Intel PRO/1000GT (82541GI)
 6:3:0: Intel PRO/1000GT (82541GI)

[r...@fw2 /usr/src]# sysctl hw.model
hw.model=Intel(R) Pentium(R) 4 CPU 3.00GHz ("GenuineIntel" 686-class)
[r...@fw2 /usr/src]# sysctl hw.ncpu
hw.ncpu=2

thanks guy's !


please, anybody with this problem ? help-me theo ! ;)

you're not providing very much information about the steps you havealready taken to troubleshoot this.

start with the simplest ruleset, and add rules under the same trafficload until you find the one(s) that make your box slow. see if you canconsolidate redundant rules into simpler ones. use tables.

-t

--
Luiz Gustavo Costa (Powered by BSD)
*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+*+
mundoUnix - Consultoria em Software Livre
http://www.mundounix.com.br
ICQ: 2890831 / MSN: cont...@mundounix.com.br

Re: [Now, other problem] Re: high load cpu with trunk+vlan+carp

Reply via email to