imap troubleshooting (was Re: martian source)

Nathan England nathan at paysonlinux.org
Thu Jul 13 15:30:23 MST 2006


I take it back, this machine is not HT. I have disabled HT on my other 
machines as in my experiences with HT the memory management was terrible! The 
machines would slow down, sometimes to a crall! But turning off HT resolved 
it and I haven't had troubles since! I would love to use it, but I just don't 
trust it. I was running SMP kernels with the HT switch in the kernel on.

I have turned off eth1 and am only using eth0.  I will try some of these steps 
to see what is happening.  The problem is dovecot just dies and I have to 
restart it. 4 times today, but since lunch it has been fine. I'm wathcing my 
logs and I don't see anything. I may setup Cyrus tonight after everyone 
leaves.

Nathan

On Thursday 13 July 2006 14:51, you wrote:
> Hi Nathan,
>
> I don't agree with your assessment.
>
> Swagging: 100 users each reading 1 100K e-mail per second = 10000K
> traffic = 10 mbps - well within a single 100mbps NIC on a saturated
> network.
>
> You can get a better idea of your NIC traffic with something like this:
> netstat -i; sleep 60; netstat -i
> take the difference and divide by 60
>
>
> I think you have a server performance issue.  You need to determine if
> it's the OS or the application or a dependence on some external server.
>
>
>
> I suggest the following:
> Remove one NIC
> Turn on HT - Why is HT turned off?
> Install package sysstat
> Install package ethereal, libpcap (and any other dependencies is may
> have) or tcpdump
>
> Review the data 1-2 days later using the sar command.  During a period
> of degraded operation:
> - run iostat
> - run vmstat
> - run netstat -an | wc -l and compare it to ulimit -n
> - run top, set refresh to 1 sec, and note the process(es) using the most
> CPU - review system logs and imap logs.  I'm not familiar with your imap
> server, but I hope it has logs.  I hope you can change the logging level to
> really verbose.
> - If possible, turn off the firewall (I've seen many problems caused by
> firewalls).
> - run ethereal on your workstation and the server and trace your imap
> traffic.  When you find lags in the traffic, investigate the cause.  Lag
> between your workstation and the server is a network issue.  Lag in the
> server can be caused by imap, name resolver, firewalls, etc.
> - You may also want to review your TCP parameters.  Your iptables script
> might have made some changes (like stuffing a 1 in
> /proc/sys/net/ipv4/conf/eth0/log_martians).  Make sure all the changes
> it makes are necessary.  I still like the idea of turning it off until
> you resolve your problem.
> - Drastic move - use strace, attach to your imap server and watch the
> output to see where it chokes (network call, file not found, permission
> denied, etc).  I was able to troubleshoot a clamav problem using this
> trick - turned out to be a directory permission issue (700 vice 755).
>
> Regards,
>
> George Toft, CISSP, MSIS
> My IT Department
> www.myITaz.com
> 480-544-1067
>
> Confidential data protection experts for the financial industry.
>
> Nathan England wrote:
> > I have looked online and found this is not a problem.
> > I decided that since my email bottle neck will most likely be network, I
> > added a second nic to the machine. They are configured as such:
> >
> > eth0 10.0.1.3 netmask 255.255.255.0
> > eth1 10.0.1.4 netmask 255.255.255.0
> >
> > route -n shows
> > Kernel IP routing table
> > Destination     Gateway         Genmask         Flags Metric Ref    Use
> > Iface 10.0.1.0        0.0.0.0         255.255.255.0   U     0      0     
> >   0 eth0 10.0.1.0        0.0.0.0         255.255.255.0   U     0      0  
> >      0 eth1 169.254.0.0     0.0.0.0         255.255.0.0     U     0     
> > 0        0 eth1 0.0.0.0         10.0.1.1        0.0.0.0         UG    0  
> >    0        0 eth1
> >
> >
> > All is well, I can point some to .3 and some to .4 specified by mail1 or
> > mail2 But Fedora is giving me the martian source error.
> > Is there a better way to do this? I have never seen a martian error like
> > this, though I have never had two nics on the same network on a Fedora
> > machine before. I've done it with other ditros and never saw this
> > error...
> >
> > Nathan

-- 
~~~~~~~~~~~~~~~~~~
Nathan England
PaysonLinux User Group 
nathan at paysonlinux.org
http://www.paysonlinux.org/

Software Development
Website Development
Linux Administration


More information about the PLUG-discuss mailing list