related
friend of mine has a startup which offers something similar: https://www.mastodonc.com/
Farmers make hay when the sun shines, and maybe data centers should crunch data and do the bulk of their work then, too. That's what the techies at HP Labs, who have been marrying renewable energy and data centers together in a testbed, are beginning to think. The HP boffins have managed to talk at least one of the recent CEOs …
Which company will have the first nuclear powered, satellite managed, submersible data center? That would make it challenging to seize. sometimes I wonder why there are still people on nuke subs anyway. Short of simply refusing to do their job, what do the crew do that a computer system can't?
In order for your data center to be of any use, it's going to need a pretty fat pipe leading to the greater Internet. Which means it's going to have to connect to a trunk line belonging to or under the jurisdiction of one or more countries (even with wireless--that data's gonna be useless unless it comes back down to earth somewhere), most of which may not take kindly to your "pirate bandwidth" and will probably compensate with high charges.
In any event, a seagoing vessel flying no colors may find itself the target of a sovereign nation which would then go on to say, "We thought it was a pirate/spy device." Unless the party that deploys it is another sovereign nation, how well would you be able to defend your right to float these around in international waters?
They also have owners who tend to monitor them since lots of people start complaining when a transoceanic link starts breaking down. Anything trying to glom onto it would likely be spotted if not by the occasional visual inspection then by the inevitable disruption to the line when you try to tap into it.
Also, if one can arbitrarily shift that much load wouldn't flattening it out allow for more efficient use of hardware?
A cynical person might take a look at that peak and think that *maybe* this is a way to sell more hardware capacity in the name of being green.
I think you missed my point... in the article "critical workload" (in blue in the demand graphs) is almost perfectly flat, 24 hours a day. I'm not a data center guru, which is why I asked the question... is that really realistic that there are no critical usage-driven fluctuations in demand during the business day? I'm not saying that they couldn't average out... just that IMHE the average production (non-idle) server has higher utilization during the day than at night.
That said, power rates are just part of the equation... the cooling and electrical systems must be sized to accommodate peak usage, which is higher (eyeballing the graph: .8 vs. .6 KWHs, so 33% higher) in the Optimal Net Zero Plan.
I'm also not sure you could just rely on flexible clock rates in CPUs to accommodate this (like I said, I've never seen baseline usage for a production server be perfectly flat over the course of a business day) but I guess it's possible.
The other issue is that to do this - which, if your goal is net zero this is really the only way to do it - they have to put in a MASSIVE solar array (not sure if my math is right, but I came up with something on the order of 50,000sf) - one that is actually oversized for peak usage to make up for nighttime. I'm not sure I'd call that "cheap power" (maybe "green power" would be a better way to describe it?). This seems to be an extreme example - which is not bad from an academic standpoint but not, I think, practical for real use.
More realistic, IMHO, you would try and average/flatline usage across the 24 hours... allowing you to most efficiently use your hardware, cooling, and electrical capacity and get as big of a solar array as you can afford (which, realistically, would still be smaller than your peak load unless you're Apple trying to drain that savings account) which would allow you to minimize your peak rate usage (commercial electric rates are, after all, highest during the day and lowest during off hours), which would further minimize costs.
This is an interesting thought experiment but neither, IMHO, efficient nor practical.
Just as one example, by my calculations, which I'm sure are not perfect, this data center being built in Utah would need to cover 1.29 square miles and cost $325 million to be Net Zero. Compared to an expected $40 million a year power bill without Net Zero, that would mean break even in 8.125 years... which isn't bad, I guess.
Why go for net zero? You'll obviously have a baseline of power use, perfect for more traditional sources, but just click on extra capacity (increase clock frequency, turn on extra cores, whatever) when extra power comes online from wind\solar\etc.
".....A cynical person might take a look at that peak and think that *maybe* this is a way to sell more hardware capacity in the name of being green." Rule One - all companies need to make a profit to survive, so you need to look at every interaction with you, the customer, from the viewpoint of "they want to sell me something to make money". The whole green bandwagon is a simple example of this - companies only started pushing green gear when they thought it would increase revenue. I'm sure the hp Lab guys involved are all lovely, but if there wasn't a way to make money off the back of the work (imagine the pitch - "buy hp cloud, we're 50% cheaper than Dell's cloud because we're energy-neutral, and green!"), then their management would be having them work on other stuff.
It would be interesting, a completely solar powered server farm, replicated around the globe in such a way that as the earth rotated and one server farm was close enough to night to produce insufficient power to operate, it hands off to the next server farm that is still producing enough power. As a farm in darkness turns back into light, it comes back online ready to resume operations.
What a joke - so to provide a 24 hour service you would probably need (at least 3) so you get 8 hours of useable sunlight per day. In a stroke more than tripling your costs (as the solar is more expensive than grid power) and increasing complexity. Well done.
Perhaps we should all realise that we need reliable, cheap and ideally low CO2 power - not unreliable / expensive solar and wind power. The irony is that for 'grid' use you need conventional power to back it up anyway.
How long will it be before the managers at the data centre get fed up of only being able to use it for 8 hours a day and just pull the rest of the power off the grid anyway as if they carry on like that they will be out of a job as their costs being 3x higher will make them completely uncompetitive.
Notice how I didn't make a single reference to it being a *good* or *profitable* idea, I just said it would be interesting, in other words it would be interesting to see how such a system may work, a great test of engineering and networking to create a system to be able to seamlessly hand off load in this kind of fashion.
Beyond a research project it is of no actual value because as you pointed out it's just not economical for numerous reasons.
As for the managers of the DC (if you read my original post I deliberately referred to a server farm, not a DC, a DC being a MUCH larger beast than a farm) getting fed up, I somehow doubt they would care that much because suddenly they are only working 8hr shifts with enough time left over for a social life, or their families. I don't know anyone who would rather be at work 16hs a day babysitting servers when they could be out living life
Sounded like a bollocks made up number. So I look around and find a report claiming energy costs are around 10% of total cost of ownership of a data centre which sounds much more realistic.
Those PV panels are not free, let's be generous and say they halve the cost of energy,
So you can reduce the cost of ownership by 5% if you only run it 8 hours a day when the sun shines which reduces productivity by 66% and limits you to work that can be done during those restricted hours.
No one in their right mind would contemplate such a rubbish trade off. I think a more interesting story is why anyone is contemplating it.
Obviously, the next step is to have the staff provide the missing power during overcast days and evenings. They will do this by riding exercise bikes turning dynamos whilst at their desks. With the move to online meetings, there is no reason for staff to leave their desks (well, nicer employers might allow them toilet breaks), guaranteeing power availability throughout the day. Vendors can then add the "feel-good" factor by claiming it's all part of a "healthy-living" scheme to reduce obesity, heart disease, etc, etc. So, green AND healthy - who could argue with that?
Of course, the Apple version will have an incompaitble powercoupling, stopping you power your non-Apple gear with your iBike.....