How Temporal Workload Shifting Can Reduce Carbon Emissions in the Cloud

Prices of photovoltaics for example has gone down to a large degree thanks to early adopters and stimulus. Now it's quite competitive without such help. But it's not the only piece of the puzzle. Other technologies deserves to be pushed towards competitiveness as well.

Also, China is not a "poor country" anymore. It's time it started setting money aside for "investment in poor countries", too, rather than build hundreds more coal power plants in the next five years.

But perhaps that is what happened. But that really just makes it even plainer that Windows is very inefficient. Why would enabling the GPU cause high power usage when the machine has been sitting idle for hours with the screen off?

Maybe, but maybe it should be what you should say? I'm consciously trying to nudge you into this direction.

> Statements like "energy is usually cleaner at 1pm than at 7pm because of solar power" absolutely hold true.

Sure. But is using more energy at those times also cleaner? I provided for a logic argument that it is only if there's surplus green energy at that time, i.e. if the need doesn't have to be satisfied by increasing use of non green energy sources during those times. After which you led me to the term of "marginal carbon intensity" which seems to express exactly this, which I'm happy with, so I said then that's the only number that matters, and if you don't have that number then currently you can't draw conclusion about the whole being greener through the shifting.

There's an argument to be made that one should move loads to 1pm since then the country could increase build out of solar energy, since there will be users of the surplus (instead of it needing storage to absorb, or left unused, or being used by more dubious consumers)! That would again be a good argument for the future, and I agree with it. But that's an argument I haven't seen you make--maybe you should.

> The goal should not be to buy more hardware and data centers are barely ever used at full capacity

Iff data centers are normally lesser used during times where the marginal carbon intensity is low, then yes, cool, no new hardware needs to be bought.

Outside of that, of course you could just ignore the need for more hardware and argue that you only shift so little load that it is insignificant compared to the reserve. But that would not be a plan for serious changes, just free riding? Maybe it's the same thinking that leads you (and others, you cite Google in the paper) used to argue that "at 1pm the mix is cleaner, hence use energy at that time": if the amount of load shift is very small, you are tempted to argue that the mix doesn't change, hence the shift is beneficial. But again, this is just ignoring an effect, since evidently if you shift a lot then that won't be true any more. I'm sure it's tempting to argue away the effect, but that's what greenwashing tries to do in general, and may be what makes a lot of people in this discussion angry. So better distinguish it from greenwashing as well as you can.

P.S.: just using up the reserve might of course be fine if moving away the shifted jobs again in case there's high non-shiftable demand coming up works quickly enough. (But it will be interesting to know how much that would be.) I don't really want to tear down your research, and some kind of simulation package to find out how much potential there is (and under what assumptions for the future etc.) would surely be cool. And I guess I've reached the point where I have to read the rest of the paper :)