We have a SunOS 5.5.1 box running Oracle 7.2 that PCs connect to. The users can submit ad-hoc queries. When they enter a bad one, i.e. one that would finish well into the 21st century, they sometimes think the PC is hung and reboot and/or power off the PC.
This query then continues to run wasting even more resources since there would not even be a way for the results to be returned. What parameters can be set using ndd so that the kernel would detect via the keep_alive packet process on the socket and clean it up? Aside from the idle timer and keep alive stuff, are there any other options in TCP/IP?
The Oracle stuff that can be set up seems to just ping the IP address associated with the socket. Unless you set this value to some very small period of time, it would only be luck that the PC was not back up and happily responding to the ICMP packets.