I'm running vmlinux-2.2.14-6.1.1smp on a Dell PowerEdge 2450. Our application connects to numerous POP3 servers on a periodic basis. We monitor the number of sockets in TIME_WAIT state, and observe this number going up and down, as expected. After a while (can be an hour or a day), we notice that sockets no longer close, and the number of sockets in TIME_WAIT state increase steadily until the kernel crashes (>5000 sockets). Is this a bug in the TCP/IP stack? If so, are there any remedies?
Created attachment 3660 [details] log of sockets in TIME_WAIT state
Are you synchronizing time on the system? _Could_ be related to #14876 and friends.
I am using ntp 4.0.99i-1 to sync the clocks amongst various servers. All other time services are off in inetd.conf.
Can you run lsof to verify the ownership of the connections? The output of netstat -a would be helpful as well. Simply knowing the number of sockets in TIME_WAIT is not enough to identify the cause of the problem. I'd also double check that the time service is actually off using netstat -a, as my blind guess is that this problem is #14876 et al as well.
I'd like to repeat the request for information...
Can you see if the one at http://people.redhat.com/teg/ helps?
*** This bug has been marked as a duplicate of 16729 ***