Tony Walsh has, as others do, some doubts about whether Second Life is sustainable as a business. But he also poses another question that I hadn’t come across before: “Is Second Life sustainable ecologically?”
He quotes Philip Rosedale, the head of Linden Lab, the company behind the virtual world: “We’re running at full power all the time, so we consume an enormous amount of electrical power in co-location facilities [where they house their 4,000 server computers] … We’re running out of power for the square feet of rack space that we’ve got machines in. We can’t for example use [blade] servers right now because they would simply require more electricity than you could get for the floor space they occupy.”
Walsh notes that on average there are between 10,000 and 15,000 avatars in Second Life at any given time, a number that’s growing rapidly. He wonders: “How much power do 15,000 human beings consume daily compared to 15,000 avatars?” Hmm. That’s an interesting question.
So let’s do the math.
If there are on average between 10,000 and 15,000 avatars “living” in Second Life at any point, that means the world has a population of about 12,500. Supporting those 12,500 avatars requires 4,000 servers as well as the 12,500 PCs the avatars’ physical alter egos are using. Conservatively, a PC consumes 120 watts and a server consumes 200 watts. Throw in another 50 watts per server for data-center air conditioning. So, on a daily basis, overall Second Life power consumption equals:
(4,000 x 250 x 24) + (12,500 x 120 x 24) = 60,000,000 watt-hours or 60,000 kilowatt-hours
Per capita, that’s:
60,000 / 12,500 = 4.8 kWh
Which, annualized, gives us 1,752 kWh. So an avatar consumes 1,752 kWh per year. By comparison, the average human, on a worldwide basis, consumes 2,436 kWh per year. So there you have it: an avatar consumes a bit less energy than a real person, though they’re in the same ballpark.
Now, if we limit the comparison to developed countries, where per-capita energy consumption is 7,702 kWh a year, the avatars appear considerably less energy hungry than the humans. But if we look at developing countries, where per-capita consumption is 1,015 kWh, we find that avatars burn through considerably more electricity than people do.
More narrowly still, the average citizen of Brazil consumes 1,884 kWh, which, given the fact that my avatar estimate was rough and conservative, means that your average Second Life avatar consumes about as much electricity as your average Brazilian.
Which means, in turn, that avatars aren’t quite as intangible as they seem. They don’t have bodies, but they do leave footprints.
UPDATE: In a comment on this post, Sun’s Dave Douglas takes the calculations another step, translating electricity consumption into CO2 emissions. (Carbon dioxide, he notes, “is the most prevalent greenhouse gas from the production of electricity.”) He writes: “looking at CO2 production, 1,752 kWH/year per avatar is about 1.17 tons of CO2. That’s the equivalent of driving an SUV around 2,300 miles (or a Prius around 4,000).”
So there you have it: an avatar consumes a bit less energy than a real person, though they’re in the same ballpark.
Mmmm, but you’re double-counting aren’t you? There’s an easy check for whether you’re double counting – you can tell because if you added all the human power consumption (in your equation) to all the avatar power consumption (etc) you’d get a figure larger than the total power consumption.
Seamus,
Sure, but it doesn’t matter in the context of this exercise. For simplicity’s sake, I decided that a single watt could be consumed by both a human and an avatar (seeing as the worlds are separate). It’s trivial because the energy consumed by humans playing Second Life has no effect on earthly per-capita electricity consumption figures (at least not yet!). Also, if you were to subtract the kilowatt-hours consumed by humans’ PCs while playing Second Life from the humans’ consumption figures (just applying them to avatars), avatar consumption would increase relative to human consumption.
Nick
Assuming the above correct, I get “running a PC (120 watts) full time is about half the energy use consumed by the average Brazilian”
(1884,000 watt-hours/(365day*24hour) = 215)
Basically, just using a PC is energy-intensive compared to the rest of the world.
Heck, the US lifestyle in general is extremely energy-intensive compared to the rest of the world:
“A typical refrigerator uses between 600 and 900 kilowatt hours per year, although an efficient one will use somewhat less (depending on the size). Refrigerators and freezers typically make up over 20% of total residential electricity consumption. Nearly all households have at least one refrigerator and about 30% own two.”
I think the language you use to describe your findings is a little misleading. You say “an avatar” – but the number 10000-15000 is the number of people online at any given time.
What you’re saying is “if you operate a PC 24/7/365 you spend more power than the average Brazilian” – which is not really very surprising.
The number of active users in Second Life is currently around 700K. So per avatar, per actual player, the power consumption is about 2.5% that of the average Brazilian – or about 40 kWh per year.
Not that that isn’t a lot of power in aggregate, but Second Life is famously inefficient per user. It’s not even the users that are causing most of the power drain but upkeep of the environment the users walk around in.
Nick – true, but I’d have thought the important thing to measure (for this exercise) was the marginal power consumption of the avatars. I wasn’t very clear when I called that double counting. So (IMHO) the equation should look more like
kWh from SL servers and aircon + (kWh from SL users’ PCs and monitors – kWh that SL users’ PCs would have consumed anyway if the users hadn’t been playing SL)
Since your equation attributes fully one third of avatar power consumption to the users’ PCs and monitors…we might assume that if SL didn’t exist all the players would be playing something else. Under that assumption the marginal avatar power consumption falls by a third. Assume instead that half of the current SL users would substitute their SL time for something that didn’t require a computer (not likely for what is still basically a geek activity) and marginal avatar power consumption falls by a sixth. Etc.
Claus, it takes one person to operate one avatar in Second Life, so “the number 10000-15000 is the number of people online at any given time” also means “the number of avatars online at any given time.” The 10k-15k stat represents peak concurrent users–SL broke the 18,000 concurrent user mark yesterday, I believe. Your description of “active users” is based on the stat “Logged In Last 60 Days.” One user could have logged in for 10 seconds and contributed to that stat.
It also bears mentioning that most client computers running Second Life are operating at peak power consumption, as the CPU and GPU are pushed quite hard. I don’t happen to have any hard data on this, however.
Claus, Tony’s right. If you want a fair comparison between humans and avatars, you need to look not at “members” or “active users” but at the actual average population of Second Life.
Seamus, You’re looking at a slightly different issue than the one I looked at, but your marginal consumption calculations are also very illuminating. Thanks.
Nick
Not sure about the server stats. 1000 new sims coming online in the land rush, apparently 30% or so increase in “landmass” (Linden quote from somewhere) suggests that 4000 is the number of sims, not servers.
There are 4 sims per server so that would 1/4 the power on the server side. Maybe I’m wrong, I don’t know. Are there 16,000 sims in SL?
I do agree that there seems to be a counterintuitive logical sidestep about the per avatar calculation- I have two accounts I log in regularly, they’re each half my SL time (well more 4/5, 1/5) so they each use less power than a single account avatar with the same time in world. This is about avatars, not people.
If there were more avatars, the energy cost per avatar on the server end would be lower per “capita”. The servers are a huge per capita overhead when logins are restricted to Lindens only, not so much when 15k are online and half the sim has offlined from grey goo :)
Big difference between avatars and Brazillians- Brazillians don’t log off and stop using energy, avatars do. Do we add the energy the typist at the chair is burning while peering at the SL client? :) It’s not about watts/minute while awake/logged in, its over the whole 24 hours. I don’t know how you measure the cost per minute of an entity that might not exist for days at a time (not logging in).
I doubt I got my own maths or assumptions right, I just don’t buy this headline.
Ace,
Follow link for source of server stat. It comes from Linden Lab’s Philip Rosedale a few days ago (and it’s actually “more than 4,000”).
Big difference between avatars and Brazillians
Damn straight. Brazilians are real, for one thing, while avatars aren’t. Perhaps I should point out that this analysis is theoretical, done for illustrative purposes.
Nick
An entertaining analysis, but the remarkable thing here is how common, entirely unremarkable things in developed countries (e.g. running a PC for many hours a day) use so much energy. Given how much energy our refrigerators use, I’m sure someone could come up with a shocking number for the amount of energy used to keep frozen burritos frozen. This would of course lead to a much less techy headline :)
With all due respect, Anthony, the ever more intensive use of computing and particularly networked computing, for entertainment as well as commerce, raises important new issues beyond the freezing of burritos. And it’s not just about another input into the high energy consumption that already marks the developed world. It’s about a whole new layer of energy consumption, which is rapidly spreading through the developing world as well. (Think of all those World of Warcraft gold farmers in China, for instance.)
The more that people become aware of the issues, the better the decisions they’ll be able to make – and the more likely it becomes that we’ll figure out and embrace innovative ways to improve the efficiency of modern computing. Dry discussions of data center kilowatt usage build some awareness, and are important, but to get more people scratching their heads and saying “hmmm, I hadn’t thought of that before,” you sometimes you need to express things in new and unusual ways. And intriguing headlines don’t hurt.
Now, I’m going to turn on my screensaver and go microwave a frozen burrito for lunch.
The calculations all look right as long as we are only talking about electricity consumption, as opposed to overall energy consumption. In Brazil, for example, energy consumption per capita is over 14,000 kWH/year, while electrical consumption is approx. the number you cited. The differences are cars, heating oil, etc. (see http://www.iaea.org/inis/aws/eedrb/data/US-encc.html) for US and other countries data in an easy to use format).
A few other tidbits:
– the amount of energy used for cooling and power distribution is probably underestimated. Many datacenters use an additional 50% to 100% for these functions (yes, those numbers are scary). They don’t make a big difference here, but thought I’d point them out.
– in the US, the average electricity/person is up around 14kWH/year.
– in the US the average CO2 emissions as a result of electricity production is a little around 1.35 lbs/kWH. CO2 is the most prevalent green house gas from the production of electricity. Your numbers come out to about 21.9M kWH/year. Producing that energy will cause 14,763 tons of CO2 to be emitted each year.
– again, looking at CO2 production, 1,752 kWH/year per avatar is about 1.17 tons of CO2. That’s the equivalent of driving an SUV around 2,300 miles (or a Prius around 4,000)
Thanks for writing this up!
Nick –
As Dave Douglas points out above, the ratio of IT equipment power consumption may be low. This turns out to be a very well researched area in fact. APC estimates (in one of their many many white papers on the topic) that in a typical data center, the IT equipment accounts for 44% of power consumption; the remainder goes to cooling (38%), lighting (3%) and the power system itself (15%).
The cooling number makes sense, as the rule-of-thumb for cooling design is that it takes 1 watt of cooling power to cool 1 watt of power consumption. When you buy IT equipment, you are generally signing up for twice its stated electricity consumption.
As far as per-server consumption, a typical mid-range server power supply will be between 500-1,000 watts. This makes sense, as average rack densities are about 1.5kW, about 10-12 servers per rack.
When you put these numbers together in your formula, now you get:
4,000 servers x 750 watts = 3mm watts / .44 for overall power load = 6.8mm watts = 6,800kW per hour
Something interesting about this. Here in Texas, where electricity is expensive I pay about $0.15 kWh (I use a renewable energy company and so pay about 1 cent more per kWh).
At the range of $0.10 – $0.13 / kWh the servers for Second Life are probably costing $700-$850 per hour to operate. Thats $500-$600 thousand per month in power charges!
It is probable that a facility gets a better rate than I do, but this is still a formidable number.
Assuming each server costs approximately $2,500 and lasts for three years, you have another $280k per month in server depreciation. With just servers and power, it may cost between $500k and $1mm per month to operate.
The Linden website shows 690,000 users logged in in the past 60 days. To get back to our original calculation, 6,800 x 24 x 30 = 4,896,000 kWh per month. Divide by 345,000 users = 14 kWh per user per month, just on the Linden side.
Interestingly, this is trivial compared to the numbers on the end-user side. A typical PC with a 200 watt power supply will consume 400 watts / hour (cooling, etc.). If left running = 400 x 24 x 365 = 3,504kWh in a year. $525 for me to run, and disturbingly over my 2,436kWh quota, for just a computer alone!
I have taken to turning stuff off around the house as much as possible.
Very interesting analysis.
John
Your headline represents a misuse of metonymy. You’re using “Avatars” in place of “Computer users.” Of course that makes the premise of this whole post more suspect.
All computer use has ecological implications, whether playing a game, shopping online, or blogging.
One might argue that most computer use is frivolous and perhaps be correct. But wither Web 2.0, not to mention the blogosphere, if frivolousness gets banned to protect the atmosphere?
How does an investor profit from these trends?
The Web 2.0 community is responsible.
See my small cartoon.
bye,
Oliver
Tony and Nick – I still have to disagree about your use of “one avatar” when you are aggregating 24/7.
Being in Second Life only accrues cost for a user when he’s actually online.
What if this was transportation economy? Would you assume of a “person” that he/she spent an entire life on airplanes wasting fuel. That number clearly would say nothing relevant about the sustainability of airflight.
But wither Web 2.0, not to mention the blogosphere
I have to say, Eponymous, you have a way with typos.
So are we gonna end up moving our household computers outside next to the heat pump?
Ace, that 4,000 server stat I quoted was spoken by Philip Linden in a video interview. It’s a number that jives with previous quotes, for example, in March, 2006 there were roughly 2,000 servers running SL.
Eponymous, where Second Life concurrent usage is concerned, “avatars” and “computer users” are the same thing. It takes a computer user to operate an avatar. The “Logged In Last 60 Days” statistic doesn’t necessarily indicate a 1:1 avatar to human ratio. Many users operate more than one avatar in a 60-day period–just not more than one avatar simultaneously on a single machine.
Claus, a user can accrue cost without actually logging in to Second Life. For example, ownership of virtual land exacts a monthly fee whether or not the owner of the land actually logs in to use the land. Perhaps I am not understanding what you mean.
Claus,
Your point about how Linden Lab derives revenue from users is pertinent to a discussion of the economics of running Second Life as a business, but my post was looking only at the question of electricity consumption and in particular comparing the electricity required by an avatar with the electricity required by a human being.
I think you’re getting hung up on the fact that one user doesn’t play Second Life continuously 24/7. But we can do the same comparison on the basis of an hour rather than a day or a year, which would reflect the experience of an actual user rather than a theoretical “living” avatar. Dividing the daily consumption by 24 gives us the hourly consumption:
avatar: 0.2 kWh
Brazilian: 0.2 kWh
So your avatar consumes as much electricity per hour as the average Brazilian consumes per hour. Which brings us back to where we started: “Avatars consume as much electricity as Brazilians.”
Nick
How does an investor profit from these trends?
Invest in hosting companies and content delivery services and pray that VCs keep pumping money into Web 2.0 startups.
So, let’s take some guess work out of these numbers. As measured by a multimeter across the leads, our power per server at 100% CPU load on both dual-core CPUs is 175 W. One server simulates 4 regions, so when we talk about 4000 regions, those are simulated by 1000 servers. So, using Nick’s math, we get: (1,000 x 225 x 24) + (12,500 x 120 x 24) = 41,400 kWh. 41,400 kWh/12,5000 = 3.3 kWh per person => 1208 kWh per person per year, which is more in line with developing nations. As others have pointed out, since this number is dominated by the residents’ PC usage, it might be more appropriate to post that using a PC consumes as much power as the average citizen of a developing nation. I think it is a little unfair to burden us with the home user’s computer, since most of those computers are on anyway, but that’s a separate discussion. The part that you are absolutely correct to focus on is that in distributed computing applications, the statistic that matters if MIPS/watt.
Thanks for the inside scoop, Cory. Philip said “4,000 machines” (which I took to mean “4,000 servers”) on the iinovate podcast, but it seems he meant to say “1,000 machines” (meaning “servers”). Ok. That dramatically changes things, obviously :)
I’ve got two notes from earlier this year–one, from Philip’s Google Talk appearance is that SL was running on 2,000 sims — the other (from eTech) is that SL was running on 2,000 CPUs. It seems clear now that it was 2,000 sims not servers.
Andrew [Linden] posted a few comments on my original post about the power consumption as well.
The first question that popped into my mind when I read this was: “How much energy are we wasting because all our software these days is written by Quiche Eaters rather than Real Programmers?”
I.e. how much energy is chewed up by the unnecessary extra memory and unnecessary extra paging to and from disc?