DSL on Linux 2.2.20: internet getting slower

DSL on Linux 2.2.20: internet getting slower

Post by Fran » Wed, 05 Feb 2003 04:59:40

Hi everybody!

I am having problems with the performance of my DSL router.

First I had the problem with Red Hat 7.3 with a 2.4.19 kernel.
I set up a new system with debian 3.0 with a 2.2.20 kernel.
THE PROBLEM was not fixed:

The speed of the internet connection is getting slower and
slower. After a re-dialin it is at full speed again. But after
a while (lets say 5 minutes of surfing) speed brakes down again.
Btw to a speed of 1kb/s. It is horrible!

My config:

debian 3.0 kernel 2.2.20
eth0 10/100ethernet card to local network (a 192.168.1.x through
     a switch)
eth1 same card to DSL modem (I am using T-DSL flat in Berlin).
using pppoe (configured with pppoe-conf).

I am doing masquerading on my local network. Even the speed of
the internet on the router machine goes down. So I know there
was a bug in the TCP/IP stack of ther kernel until 2.2.12 but
that cannot be the problem because I am using a newer one.

ifconfig does not count any collisions in network traffic.

I am really out of ideas!
Any suggestions for me?

Greetings Frank


DSL on Linux 2.2.20: internet getting slower

Post by Fran » Wed, 05 Feb 2003 19:20:29

I think I am getting closer...

Everytime the connection to the internet gets very
slow (for the MASQ local net) the collision count
shown by ifconfig for the eth1 interface (that one
connected to the DSL modem) is growing.

But what the *! What can I do now? The dmesg says
that eth1 is in half duplex, like it should!
Why are there collisions after a period of being
on-line? And not at the beginning?

Every help appreciated.



1. demand DSL with kernel 2.2.20 modules


(hoping that this is the right group)

I am using debian 2.2r5 and i want demand dialing with DSL. Not with
rp_pppoe but using the kernel modules.

So i read some FAQ's and do it as they said:
- build pppox and mssclampfw as kernel modules and the pppoed
(pppoe version 0.4 build 20001008)
- patched the pppd (v.: 2.3.11) with the demand.c.patch found at

This works fine if I dont use demand dialing. But if i try the demand
option for pppd i can only dial in once. The next time I try to dial i
get the error "LCP: timeout sending Config-Requests". With debug i get
"sent [LCP ConfReq id=0x3 <mru 1492> <asyncmap 0x0> <magic
0x9b491887>]" messges but this helps me not much.

After one dialout the default route is set to the ppp0 interface.

Feb 22 12:55:23 cavalieri pppd[333]: sent [LCP TermReq id=0x2 "Link
Feb 22 12:55:23 cavalieri pppd[333]: rcvd [LCP TermAck id=0x2]
Feb 22 12:55:23 cavalieri pppd[333]: Connection terminated.
Feb 22 12:55:23 cavalieri pppd[333]: Connect time 0.8 minutes.
Feb 22 12:55:23 cavalieri pppd[333]: Sent 1940 bytes, received 2554
Feb 22 12:55:24 cavalieri pppd[333]: Script /etc/ppp/ip-down finished
(pid 344), status = 0x0
Feb 22 12:55:35 cavalieri pppd[333]: Starting link
Feb 22 12:55:35 cavalieri pppd[333]: Serial connection established.
Feb 22 12:55:35 cavalieri pppd[333]: Connect: ppp0 <--> /dev/pppox0
Feb 22 12:55:36 cavalieri pppd[333]: sent [LCP ConfReq id=0x3 <mru 1492>
<asyncmap 0x0> <magic 0x36b04ab7>]
Feb 22 12:56:03 cavalieri last message repeated 9 times
Feb 22 12:56:06 cavalieri pppd[333]: LCP: timeout sending Config-
Feb 22 12:56:06 cavalieri pppd[333]: Connection terminated.

My options file looks like that:
connect /bin/true
idle 960

mru 1492
mtu 1492
lcp-echo-interval 60
lcp-echo-failure 2
#kdebug 7

Does anyone have that way running or can anyone tell me what the problem
Thanks in advance...

        [dont send me an email]

2. New To Linux

3. poptop on Linux 2.2.20

4. Cloning Solaris x86 boot drive

5. Linux 2.2.20 -> ps -u floating point error.

6. Upgrading Imlib's

7. Linux 2.2.20pre11

8. Installing Ethernet Cards w/o X

9. Linux 2.2.20pre7

10. DOSEMU 1.0.2, Slackware Linux (Kernel 2.2.20) and "DOSNET"

11. Compaq Gigabit Ethernet NC6136 (fiber, NOT coper) on linux 2.2.20

12. scsi errors on debian linux 2.2.20-pmac on Unax C600

13. Linux 2.2.20pre6