Re: TCP Stall

bofh@snoopy.virtual.net.au
Sun, 06 Apr 97 12:11:36 +1100


>I have now found the EXACT sequence of events that leads to the TCP Stall.
>It is repeatable. I don't know WHY it happens, however I now know WHAT is
>happening. The event that leads to the stall is the first line in the
>attached file.

>I enabled the character generator (chargen, port 19) so I have a
>continuous stream of characters. The remote (slow-speed) end of the link
>executes:

># telnet chaos chargen

>A process on my ppp server, that can listen on the Ethernet backbone
>executes:

># tcpdump -nlpS -tt host chaos >tcp.dmp

>It looks as though a cache-entry has expired, the host sends an ARP
>request to find the hardware address of the already established client.

>The host gets very confused but eventually continues to send data 31
>seconds later.

>So the new questions are; "Should an ARP cache entry expire on an
>established connection?", and; "If this happends, what bad things can
>happen?".

You've apparently found/experienced a different TCP problem to me. My problem (on which I've posted a message including tcpdump output merely 5 minutes ago) occurs between 2 Linux systems that are directly connected via a PPP link. So whatever my problem may be it's definately not ARP and may not be directly related to what you're experiencing.

My problem is also very repeatable. On my Linux server I have a partial mirror of ftp.kernel.org. When I FTP the latest Linux kernel source I get a pause in the transfer every few hundred K throughout the transfer.

Russell Coker