I'm running an Ubuntu 11.04 server release with kernel version 188.8.131.52-server, and for as long as I've had the system it has hung randomly. I'm starting to look into the LSI MegaRaid controller, but was wondering if anyone else experienced similar. Here's the system essentials:
- Motherboard: Supermicro H8SGL
- CPU: AMD Opteron 6172 12-core
- RAM: 64GB DDR3-1333 ECC
- LSI MegaRaid MR9240-8i controller, with 8 SATA drives attached in RAID-5 with 2 hot spares - 4TB total available storage
I haven't done an intensive memory check yet, so that's still on the table. The system runs for about a month before up and choking to death for no apparent reason. kern.log sheds no light. No info from dmesg. Leaving a monitor attached with screensaver turned off and kernel messages dumping out there also provides no useful information. I've thought about upping the kernel diagnostic output, but am unsure how to achieve that without recompiling the kernel. The machine is intended to be a VM host, so you can appreciate that this is not really an acceptable level of service when running 5-15 virtuals! I've been using KVM as the underlying hypervisor, and it's run perfectly fine on lower-end (desktop-class) machines - everything from AMD X2 multicores to Core2-Quads. All of them are running various incarnations of Ubuntu, usually 10.04 or better. The system freezes used to happen more frequently, but decreased a bit with updates and also with going from 10.04 to 11.04.
As far as the RAID card is concerned, I only suspect that because it wouldn't be the first LSI card to give me similar/major issues (even a "reputable" vendor such as LSI is not without its bad devices - I usually use mdadm anymore, just because it's safer and runs just as good). Unfortunately the entire system is on the RAID controller, so migrating it off will be tricky but doable. The system is currently only running 5 virtuals now, and I'll probably migrate the rest of them over to other hosts until I can get this machine straightened out, which means I can do an intensive memory scan in the meantime. But any insight or information anyone else has to offer, or instruction on enhancing the output of kernel diagnostics would be great! I'm also willing to hook up a serial cable to a nearby machine to capture data that way, but I've never done it before and would appreciate a link if someone has one handy AND thinks such a thing would be useful.