IT, and especially Operations, collects of a lot of Usage Data. Most of this Usage Data is trended and used in capacity planning and debugging. It’s useful internally and probably not that interesting externally.
A lot of the more interesting data comes in the form of bandwidth graphs, NetFlow analysis, DNS statistics, load balancer and router performance graphs, China bandwidth and China GSLB traffic distubution.
We’ve recently begun trending power usage from Mozilla’s primary data center in San Jose1. The following graph shows wattage per row2:
The genesis of this was inspired by Nobel Laureate Al Gore. I had the opportunity to see Mr. Gore give the keynote at the RSA Conference in San Francisco last April. His comment, “Make the invisible, visible,” resonated and stuck with me.
On average we’re using about 50kW (and remember, that’s a point-in-time measurement, the amount of power we’re using at any given time). I find that number to be too abstract — I’ll try to put that number into more meaningful terms (this calculator was helpful):
- If the local utility were to bill us for this, we’d get a bill for about 36MWh. For perspective, my PG&E bill for last month was for 484kWh.
- The Tesla Roadster could go about 370km (230 miles) with 50kWh.
- In California, 1 kWh of electricity generates 0.878707 lbs of CO2 emissions. Our San Jose data center generates 43.94 lbs of CO2 (or 0.02 metric tons). For comparison, an average car (20 mpg) generates 44 lbs of CO2 after driving about 45 miles.3
- According to the DOE, the average U.S. home uses approximately 936kWh of electricity per month, which is about 31 kWh/day. Daily we consume 1.2MWh. At that rate we could power ~38 homes every day.
- Or, with roughly 250 million Firefox users, 50kW is 200 milliwattsmicrowatts/user (.2mW) (or 144Wh/month).
That last one is an interesting metric to measure power usage. It’s more a measure of efficiency, efficiency per user.
I’m left with some questions:
- Who else is showing this sort of energy usage data?
- Is this a useful metric? Can I use this to compare with others, apples-to-apples? Does it mean anything?
- Is Mozilla’s data center energy usage good? Bad?
- How can we use this data to help drive business decisions?
- Does energy consumption change during Firefox releases?
Lastly, I want to point out that the graph on this page is a link to a live image. The data for this graph is pulled every five minutes and the graph image is updated twice an hour. At some point this blog post will get lost in my archives so I’ve created a Wattage page with this graph and more historical graphs. As we’re able to, I’ll add data from Mozilla’s other data centers.
1 Admittedly, this graph under reports the true energy impact. It doesn’t take into account the amount of energy required to cool the data center space we’re in and is obviously missing both the Amsterdam and Beijing data centers.
2 This per-row graph shows something interesting.
16/101 has 9 racks and 212 devices in inventory.
16/102 has 6 racks, two of which are just patch panels. All the routing and load balancing hardware is in this row and inventory shows 59 devices.
16/103 has only 7 racks and 231, the most dense row and from the graph, less power hungry than
16/101. It’s also the row with all 141 Mac Minis.
3Calculations provided by Mira @ TerraPass.