I'm running Red Hat Enterprise Linux AS release 3 (Taroon Update 5) on a Dell server with openafs and local filesystem.
It is connected to a fairly large network with red hat linux work stations and solaris machines on it.
All run a mix of openafs and local file system.
Authentication is made with kerberos against a server residing in the afs portion of the file system.
Many students log in to the server via ssh for various reasons and if they don't log out properly but only close down the terminal window they used to connect to my server there will be a number of processes that hang.
These processes cannot be contacted and they don't die. They are allways in pairs, one parent process that belongs to root and one child process that belongs to the person that logged in. If I try to kill them there is no respons from the parent processes and the child processes become "defunct". If there are more than five such process pairs sshd waits forever when someone makes a new login attempt. It works fine to login with kerberised telnet. I cannot reboot by the command "reboot", it hangs too - probably waiting for sshd that will not shut down. I can, however, issue a "init 6" which reboots the server.

I found a bug report and used the suggested remedy - issue a
shopt -s huponexit
in /etc/bashrc

and
ClientAliveInterval 15
ClientAliveCountMax 4

in /etc/ssh2/sshd2_config to make sshd kill clients after a minute if they don't respond to "are ou alive" questions.
It still doesn't work and I'm somewhat new in this field. Can someone suggest a solution?

Does 'kill -15' make any difference?

Not many things seem to work. I have decided (for now) to allow login only via kerberized telnet. It works without problems.