Wes Shull writes:
On 4/26/06, Andrew Haley <aph(a)redhat.com> wrote:
> #include <netdb.h>
[snip]
I don't think Andrew's code is showing the real problem here, because
it's not actually trying to send anything over the connections after
they've been opened, to make sure they're still alive.
At first I thought Hendrik was on crack, but I'm able to reproduce
something similar on my FC5 system. It's not related to the open file
ulimit.
I hacked up the Tcl code at the bottom to open a bunch of sockets,
connect to itself, and pass data back and forth; run it and see what
you get. On my 2.6.16-1.2080_FC5 system, all but 19 of the
connections die after the client side sends some initial data. I'm
thinking that some connection tracker (i.e. iptables/netfilter) is
losing track of them and so kicking back a reset.
It works for me, bouncing data back and forth. What does it do on
your box? What is the output?
Andrew.