I have a 2.1.128 box connected via PPP to a 2.0.35 box. Today I was
hammering quite heavyly the 2.1.128 box and as the system load went
up I started to see that packages transmitted over the PPP link
started to get lost.
I was torturing the 2.1.128 by doing "dd if=/dev/hdc of=/dev/null"
(where /dev/hdc is my CD-ROM) and at the same time a "find
/usr/src/linux -type -exec cp {} /dev/null \;". This seems to generate
an interrupt storm (/usr/src/ is mounted on ide0, and obviously,
/dev/hdc is on ide1).
To test the PPP link I pinged the PPP host on the other side (the
2.0.35 box) and got this:
41 packets transmitted, 10 packets received, 75% packet loss
even though I saw the modem lights flashing for both transmit and
receive (this leads me to think that the TCP stack was indeed sending
the ICMP packets, the remote host was responding to the ICMP echo but
the TCP stack was not processing the recived packets).
As soon as I killed the processes that were drowning the system the
PING's started to como back and forth with no packet loss.
This is a Pentium 75 box so I guess you won't be able to reproduce
this if you have a lot of computer power and fast hardware.
My question is: is this the normal behavior for a system that is
being heavyly hammered like this? Am I pushing this hardware to its
limits, or there is somekind of problem in the TCP stack?
Thanks!
Eloy.-
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/