We are currently running into an issue on a RHEL 5.6 server. The server randomly becomes unstable, kernel wait messages are showing up in /var/log/messages, and we've now had it reboot twice. We've sent all of our available information to Red Hat, but to troubleshoot further they have stated that they need kdump and kexec set up which includes the following:
- Enable the kdump service in chkconfig
- Set hung_task_panic to 1 in /etc/sysctl.conf
- Add crashkernel=128M to grub.conf to set up a memory resident monitor for kernel dumping
My big concern is the memory resident aspect of this. Has anyone else enabled kernel core dumping before? Was there any major performance impact from the memory resident? I'm just looking for feedback on others experiences.