Hanni Ali wrote:
> 
> I would be interested to hear anyone else's experiences with IPMI.
> 
> Regards,
> 
> Hanni

I've recently started using IPMI software (ipmitool, ipmievd) where I 
can, but mostly for monitoring disk arrays, temperatures, and things 
like that that aren't available on some systems (nearly all in my case) 
via lm_sensors, ACPI, SMART, etc..  Support for the various chips is not 
great, but its nice to get alerts that the BIOS thinks a disk is dead 
before the other one dies or someone finally arrives in the room to hear 
some server beeping.  That can be a pain to track down.

As for the HW level interaction with them security and IP usage is often 
a problem.  Ideally you could use the second NIC which is generally 
available on most servers to setup a separate network for IPMI 
management, but again that's not especially easy to do in my case.  So 
far my interactions with it have been limited to power cycling a machine 
from home, but I haven't gotten console access to work, so unless you 
can do that if a kernel upgrade goes bad you might be SOL in that case. 
  There are probably fancy tricks you could do with grub, but I haven't 
gotten that far into it.

I think network booting would probably be your best bet.  In a cluster 
setup machines should only take on one of a couple roles and it would be 
easier to manage a handful of images rather than individual machines. 
Of course you also need to figure out redundancy for your network image 
server at that point.

VMware ESX w/ VMotion might also be a nice way to handle all of this, 
but that doesn't totally get around the issue since a physical server 
could have problems taking all of your virtual servers down and you'd 
still need to rely on IPMI or serial console to address that machine. 
It also doesn't help the initial cost problem.

Curious to see what you go with.

Brian
-- 
gentoo-cluster@g.o mailing list