- ecological: reduce the environmental impact of the activity
- financial: drive down energy cost
- availability: make sure a power circuit is not about to trip
- scalability: know how many servers you can add on a circuit, in a rack or a datacenter
- troubleshooting: detect incident like A/C issues
Some basic rules I always recommend:
Even if you have a team on site 24/7, always use manageable power bars, that will allow you to track the power consumption of your equipments and power then up and down remotely.
Configure your bars so that all the servers are not powered up at the same time if the power happen to go down and up. A "boot surge" could trip the circuit even if you are far from the limit when the systems are operating. You can also use this delay between the servers startups to handle servers/services dependencies (boot a file storage first for example).
Never exceed 70% of the amperage of a circuit. Even if the power consumption appear very stable, servers needs much more juice to cool down when the A/C fails.
Understand how your servers behave on an energy standpoint when booting, when rebuilding a RAID container, when the CPU load is high, when the temperature raises...
Recently, I have discovered a $29.90 tool that helped me a lot: Kill-A-Watt.
Connect a server to this device and stresstest it. You'll immediately see the Volts, Amps, Watts, Hz and VA.
Posts
