Linode.com Forum

[Squishy]

Hi,

I was wondering if anyone in the know could give me some hints about how much power a Linode VPS (say a 360 meg one) consumes.

I'm interested in doing a back-of-the-envelope calculation of carbon footprint (which will probably be completely off...)

I suppose knowing the average consumption of a single physical machine would be a good starting point, unless this is classified information -- :wink: -- or the info's just not available.

Thanks! :)

[caker]

Our host servers average around 2.5A @ 120v each. Watts = Volts * Amps, sot that's 300 Watts per host.

http://www.linode.com/faq.cfm#how-many-linodes-share-a-host

A single Linode 360's share of its host's power consumption would be about 7.5 watts.

-Chris

[Squishy]

Brilliant, thanks!

So assuming electricity consumption of 0.0075 kW, that's

0.0075 * 24 * 365 = 65.7 kWh in one year.

According to the calculator below, that's about 0.03 tonnes:
https://www.climatecare.org/calculators/business/

Still, I suppose there are lots of other factors to consider, like the source of the electricity, power usage of visitors to the server, manufacture and disposal of machines, etc.

Cheers again

[pclissold]

caker wrote: A single Linode 360's share of its host's power consumption would be about 7.5 watts.
The data centre probably uses about half that again for air conditioning to pump the heat out of the server room.

[bdonlan]

pclissold wrote: caker wrote: A single Linode 360's share of its host's power consumption would be about 7.5 watts.
The data centre probably uses about half that again for air conditioning to pump the heat out of the server room.
More than half, no heat pump is 100% efficient.

[Stever]

bdonlan wrote: More than half, no heat pump is 100% efficient.
Taken thermodynamics lately? Heat pumps can very easily move more energy than they consume.

EDIT:In case my rude assertion ;) isn't convincing:
http://en.wikipedia.org/wiki/Heat_pump#Efficiency

[vca]

Another way of looking at this is to examine the case for replacing a machine that's on 24x7 (either at home or at the office) with a linode of some size and see if you can justify it on the basis of power cost savings alone.

Consider, for simplicity an electric cost of $0.10/kwh (my current rate is $0.093/kwh plus additional fees that mean its really $0.13/kwh).

Then for a machine that consumes X watts of power, the monthly cost (for simplicity assume a month has 30 days) is:

cost = 30 * 24 * X/1000 * 0.10

so for a cost of $20/month (i.e. the price of a linode) you would have to replace a machine that consumes:

X = 1000 * cost/(30*24*0.10)

X = 1000 * 20/(30*24*0.10)

X = 278 W

Now that's a pretty big machine (in fact about the same size as the actual machine that hosts the linode).

However, your typical desktop box will draw in the range of 100-150W when running idle, which amounts to about $10.00/month in raw electric costs if left on 24x7.

So how do you justify the other $10/month?

Well, a typical computer depreciates to worthless over a 3 year (i.e. 36 month) period, so if you take the initial purchase price and divide it by 36 then that's what you are paying per month to use the machine. So for a low-end box at $400 that would be $11/month.

Neat eh? A $20/month linode is an economic replacement for a $400 PC that needs to be left on 24x7.

For a lot of applications a $10/month linode would be even easier to make the economic argument work, and if administration costs are too much on a monthly basis perhaps offering this only on a pre-paid quarterly or yearly basis would be the way to go.



Regards,

Stephen

[mwalling]

vca wrote: So how do you justify the other $10/month?

What about internet connection? Assuming your Linode is a web server/mx, how much per month does it cost to get a reasonable sized internet connection (a symmetrical one)? I know Time Warner Cable's Road Runner is $45/mo for 10Mb by 1Mb, and running "servers" on their residential network is a violation of their ToS, so technically you would have to go up to one of the small business packages.

[Xan]

Hang on there, fellas. I think Caker's wattage calculation is overly simplistic.

This is alternating current, so the apparent power is V * A. But the real power is V * A * pf, the power factor.

For DC, the power factor is always 1, as well as for a purely resistive load. A computer is not purely resistive. I'm guessing the server's power factor is about .7, but that really is just a guess. The folks who run the data center could probably tell you better.

If that is about right, it would take the actual energy use of a server to 210W.


I'll save the conversation about carbon dioxide not being a pollutant, and about all this "carbon footprint" talk being a load of hot air, for another time. :-)

[vca]

mwalling wrote: vca wrote: So how do you justify the other $10/month?

What about internet connection?

Assuming your Linode is a web server/mx, how much per month does it cost to get a reasonable sized internet connection (a symmetrical one)? I know Time Warner Cable's Road Runner is $45/mo for 10Mb by 1Mb, and running "servers" on their residential network is a violation of their ToS, so technically you would have to go up to one of the small business packages.

True I did leave that out under the assumption that you'd still need to have a DSL/cable connection for your home and that you had been running your 24x7 device off it in a probable TOS violation (for example you might have been running a small web server to share photos with other family members). So you're not saving anything here by going the linode route.

That said, if you had been paying the premium to turn a home connection into a "business style" connection (so that you were legit on the TOS and/or to get a static IP address), then the switch to linode is a complete no-brainer as the differential (at least for me) between home and business pricing is about $50/month - which buys quite a bit of linode bliss.

[Stever]

Xan wrote: I'm guessing the server's power factor is about .7, but that really is just a guess.

Their retired servers had PFC power supplies, so I'm guessing the new ones do as well. That would put the pf up close to one and Caker's estimate would be pretty much correct.

http://www.linode.com/forums/viewtopic.php?t=2877

This is what happens when you mix computer geeks with electronics nerds ;)

[Xan]

Quote: Their retired servers had PFC power supplies, so I'm guessing the new ones do as well. That would put the pf up close to one and Caker's estimate would be pretty much correct.

That would make sense. I'm sure the datacenter much prefers machines with power factor correction!

[mwalling]

Spectrum ran an article at one time about a plan to wire data centers with DC right to the machines, instead of each machine having its own PSU... What ever happened to that idea?

[Jay]

mwalling wrote: Spectrum ran an article at one time about a plan to wire data centers with DC right to the machines, instead of each machine having its own PSU... What ever happened to that idea?

That's actually what google does.

[Xan]

mwalling wrote: Spectrum ran an article at one time about a plan to wire data centers with DC right to the machines, instead of each machine having its own PSU... What ever happened to that idea?

IIRC, the energy savings aren't quite as spectacular as one might expect. It doesn't make sense to run two different lines, 5V and 12V, plus the lower the voltage, the more juice you lose on the lines. I think they were talking about running 60V or 48V or something. Then you still have to transform it to do anything with it. Granted, it's better than converting AC to DC, but maybe not so much better that it's worth redesigning your datacenter.