[ale] uptime, cool--New topic, keeping power on

[Dow Hurst]

Mark,
We keep all our machine on 24/7 since people log in and out using the
workstations as compute nodes.  We keep everything on a UPS of some kind
to smooth the power and minimize down time.  Typically it is only kernel
related security updates that require a reboot.  I have ~55 machines of
varying OS GNU/Linux versions I admin and none require regular reboots. 
Your biggest problem with not rebooting is disk related aging for lower
quality disk drives.  Your biggest caveat with rebooting is amperage
overload on the circuitry.  You do have to balance power protection and
power costs if your going to keep stuff on all the time.

APC UPSes are supported well by the apcupsd package.  The machine can
know what the UPS is doing and respond to power outages monitored by the
apcupsd daemon.  If you have a less well supported brand the package NUT
is designed to work with any UPS.  I've heard that New York city has
such good power in Manhattan that only voltage regulation is needed.  I
don't know how true that is but a large research cluster ~250CPUs was
installed at Cornell with only voltage regulators of high quality. 
Maybe a UPS was used on the NFS file server.
Best wishes,
Dow