Disaster under heavy network load on 2.4.x

Disaster under heavy network load on 2.4.x

[Michal Margula]

Hello!

My friend told me to noticed you about problems I had with 2.4.x line of
kernels. I started up from 2.4.3. Under heavy load I was getting
messages from telnet, ping, nmap "No buffer space available". Strace
told me it was error marked as ENOBUFS.

First thought was it was my fault. I asked many people and nobody could
help me. So I tried 2.4.5. It was a disaster also (should I mention few
oopses?:>).

Second thought was to try 2.2.19 and it was good choice. Now there are
almost no messags like those above. Only thing that still happens is
"Neihgbour table overflow".

Some data about my Linux box:

2 x PIII 800 MHz/1024 MB; 2 x Intel EExpres 100; 3 x 3com 3c900B-Combo.
Summarizing all traffic about 5mbit at the moment.

# arp -an | wc -l
   1018

Any more info needed?

PS. It would be nice to be CCed with replies, beacause I am not
subscribed to LKML.

-- 
Michal Margula, alchemyx@uznam.net.pl, ICQ UIN 12267440, +)
http://uznam.net.pl/~alchemyx/, Polish section of Linux Counter maintainer

Re: Disaster under heavy network load on 2.4.x

[Martin Josefsson]

On Mon, 11 Jun 2001, Michal Margula wrote:

> Hello!
>
> My friend told me to noticed you about problems I had with 2.4.x line of
> kernels. I started up from 2.4.3. Under heavy load I was getting
> messages from telnet, ping, nmap "No buffer space available". Strace
> told me it was error marked as ENOBUFS.
>
> First thought was it was my fault. I asked many people and nobody could
> help me. So I tried 2.4.5. It was a disaster also (should I mention few
> oopses?:>).

Would you mind running those Oopses through ksymoops and send the
backtraces to this list?

/Martin

-- 
Linux hackers are funny people: They count the time in patchlevels.

Re: Disaster under heavy network load on 2.4.x

[David Ford]

It looks like you don't have 'lo' configured, i.e. your 127.0.0.1 interface.

David

Michal Margula wrote:

>Hello!
>
>My friend told me to noticed you about problems I had with 2.4.x line of
>kernels. I started up from 2.4.3. Under heavy load I was getting
>messages from telnet, ping, nmap "No buffer space available". Strace
>told me it was error marked as ENOBUFS.
>
>First thought was it was my fault. I asked many people and nobody could
>help me. So I tried 2.4.5. It was a disaster also (should I mention few
>oopses?:>).
>
>Second thought was to try 2.2.19 and it was good choice. Now there are
>almost no messags like those above. Only thing that still happens is
>"Neihgbour table overflow".
>
>Some data about my Linux box:
>
>2 x PIII 800 MHz/1024 MB; 2 x Intel EExpres 100; 3 x 3com 3c900B-Combo.
>Summarizing all traffic about 5mbit at the moment.
>
># arp -an | wc -l
>   1018
>
>Any more info needed?
>
>PS. It would be nice to be CCed with replies, beacause I am not
>subscribed to LKML.
>

Re: Disaster under heavy network load on 2.4.x

[Andi Kleen]

Michal Margula <alchemyx@uznam.net.pl> writes:

> Hello!
> 
> My friend told me to noticed you about problems I had with 2.4.x line of
> kernels. I started up from 2.4.3. Under heavy load I was getting
> messages from telnet, ping, nmap "No buffer space available". Strace
> told me it was error marked as ENOBUFS.

You probably need to increase various sysctls to handle your load. 
E.g. /proc/sys/net/ipv4/tcp_mem, net/core/[rw]mem_{max,default}, neigh/
defaults/gc_thresh* (for the ARP table) 


> 
> First thought was it was my fault. I asked many people and nobody could
> help me. So I tried 2.4.5. It was a disaster also (should I mention few
> oopses?:>).

When you have oopses run them through ksymoops and send them to the list.

 
-Andi