hardware problem ?

[Net Llama!]

On 01/18/04 16:03, Vu Pham wrote:

> ----- Original Message ----- 
> From: "Net Llama!" <netllama at linux-sxs.org>
> To: <linux-users at linux-sxs.org>
> Sent: Sunday, January 18, 2004 12:36 PM
> Subject: Re: hardware problem ?
> 
> 
> 
>>On 01/18/04 10:23, Vu Pham wrote:
>>
>>
>>>This morning my  server ( RH 7.2  kernel 2.4.20 ) suddenly failed its
>>>networking functions. This server has 3 network cards eth0, eth1, and
> 
> eth2.
> 
>>>I still could login to it. top and ps does not show me anything strange
>>>except this line
>>>root       786     1  0 11:25 ?        00:00:00 [eth2]
>>>Restarting /etc/initt.d/network did not help.
>>>
>>>>From the server, I could not ping or telnet to any other site.
>>>>From other workstations, I could not ping to it or telnet to it.
>>>
>>>The only network operation I could see it from the outside is ARP. After
> 
> I
> 
>>>pinged unsuccesfully that server from my workstatiom, I can see its MAC
>>>address on my workstation.
>>>
>>>What does that process [eth2] mean ?
>>>
>>>Any tips that I can find the problem ?
>>
>>Have you checked dmesg or messages for errors?
> 
> 
> Yes, the first thnig I did was to I check  both of them. In fact, there were
> a lot of  messages like :
> -----------------------
> Jan 18 10:40:15 pluto kernel: ip_conntrack: table full, dropping packet.
> Jan 18 10:40:15 pluto last message repeated 9 times
> Jan 18 10:40:22 pluto kernel: NET: 4 messages suppressed.
> Jan 18 10:40:22 pluto kernel: ip_conntrack: table full, dropping packet.
> -----------------------
> 
> but *mixing* with these messages, there were still some log of FTP
> transactions from other workstation to this server. I mean after message of
> "dropping packets" there were some other successful transactions for ftp.
> 
> Could it be some problem with iptables ?

No, the problem is the very large number of tcp/ip connections hitting 
your box.  You need to increase the number of connection tracking table 
entries.  You can review your table with:

# cat /proc/net/ip_conntrack

The max number of connections is set in

# cat /proc/sys/net/ipv4/ip_conntrack_max

You can increase it with:

# echo "some_number" > /proc/sys/net/ipv4/ip_conntrack_max

Did the output of ifconfig for each interface show any errors?  What 
kind of NIC(s) do you have?  There are some cases of the old eepro100 
driver hanging under very heavy network load traffic (like you 
apparently have).

-- 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
L. Friedman                       	       netllama at linux-sxs.org
Linux Step-by-step & TyGeMo: 		    http://netllama.ipfox.com

  16:15:01  up 42 days, 21:00,  1 user,  load average: 0.02, 0.07, 0.12