Timeout too long if no network

I have a 2.4.18 kernel based system that I use in two configurations. In
the first, the system boots up connected to a LAN, and gets an IP via
DHCP and mounts a remote filesystem. In the second scenario, the system
boots with no LAN connection, and mounts its own filesystem from disk.

It works for both right now, but it takes 2 minutes or so to boot up
without the LAN. The problem is there are two long timeouts at "Sending
dhcp requests ....." and "Reopening network devices.....".

Is there a way to shorten these timeouts to 10secs or so, or at least
turn networking on and off easily via some config file?

I noticed there are quite a few network related options:
#/sbin/sysctl -a

tns1 wrote:

> I have a 2.4.18 kernel based system that I use in two configurations. In
> the first, the system boots up connected to a LAN, and gets an IP via
> DHCP and mounts a remote filesystem. In the second scenario, the system
> boots with no LAN connection, and mounts its own filesystem from disk.
>
> It works for both right now, but it takes 2 minutes or so to boot up
> without the LAN. The problem is there are two long timeouts at "Sending
> dhcp requests ....." and "Reopening network devices.....".
>
> Is there a way to shorten these timeouts to 10secs or so, or at least
> turn networking on and off easily via some config file?
>
> I noticed there are quite a few network related options:
> #/sbin/sysctl -a
>
Also when the LAN is present, there is a long wait for:
"Looking up port of RPC 100005/1 ...."
This looks like another unnecessarily long timeout.

> Is there a way to shorten these timeouts

Hi,

I use Debian and always got annoyed too by the few seconds wasted for
dhcp to finish. Well on my laptop it's just a few seconds and not minutes
;-)
On debian the package dhcp-client [1] is installed by default. Replacing
dhcp-client with pump [2] seems to have helped a little. However, I
disabled the automatic network configuration for the interface eth0 and
created a simple script which I placed in /etc/init.d/pumpeth0.sh in the
end.

--- Script: pumpeth0.sh ---
#!/bin/sh
/sbin/pump -i eth0 >/dev/null 2>&1 &
echo "Backgrounding DHCP configuration for interface eth0"

To complete you just have to make the appropriate symlinks in /etc/rcX.d/

Regards,

Daniel


[1] http://www.isc.org
[2] This is the DHCP/BOOTP client written by RedHat.

Daniel Stirnimann wrote:
>>Is there a way to shorten these timeouts
>
> On debian the package dhcp-client [1] is installed by default. Replacing
> dhcp-client with pump [2] seems to have helped a little. However, I
> disabled the automatic network configuration for the interface eth0
>
Where specifically did you disable this?

I was hoping I could leave everything enabled and just shorten the
delays (timeouts) by some ifconfig parameter or conf file entry.

> Where specifically did you disable this?

I just commented the two eth0 entries in /etc/network/interfaces out.
This is debian specific and propably different on other linux
distributions.

> I was hoping I could leave everything enabled and just shorten the
> delays (timeouts) by some ifconfig parameter or conf file entry.

You have to look in the dhcp client configuration file.
For pump you can define the timeout in /etc/pump.conf

There is propably a similar setting for the dhcp client you are using.

Regards,
Daniel

>Long delay during "Sending dhcp requests ....."
>Also when the LAN is present, there is a long wait for:
>"Looking up port of RPC 100005/1 ...."

>>I was hoping I could leave everything enabled and just shorten the
>>delays (timeouts) by some ifconfig parameter or conf file entry.
>
> You have to look in the dhcp client configuration file.
> For pump you can define the timeout in /etc/pump.conf

After long and tedious greps, I found the DHCP timeout in (surprise)
ipconfig.h . There are a bunch of timeout constants, but it looks like
the defaults are set to 2 reopens, 6 tries with exponential backoff that
roughly doubles the timeout each try. This looks like a worst case
network configuration. I should be able to trim this way back and still
get a connection with:
CONF_OPEN_RETRIES can be lowered from 2 to 1
CONF_SEND_RETRIES can be lowered from 6 to 3

Sadly, these are only changed by re-compiling. I'll get to it later.

Finding the RPC timeout has proven too convoluted for me to follow.