Re: [Userops] Why is it hard to move from one machine to another? An analysis.

[韋嘉誠]

On Fri, Apr 10, 2015 at 12:42 PM, Elena ``of Valhalla''
<address@hidden> wrote:
> On 2015-04-09 at 20:14:16 -0400, Dave Crossland wrote:
>> > Even worse, that model wastes energy at a spectacular rate.
>> (Sure, its industrial scale so it uses a lot of energy. Running things at a
>> personal scale necessarily can't use much. But, the economies of scale mean
>> I suspect its not actually more wasteful watt for cpu cycle than your mini
>> pc, plus, acmecorp's solar/dam powered datacenter in the desert/forest is
>> greener than your mini pc powered by the utility grid which runs on
>> coal/gas/nuclear.
>
> using watt per cpu cycle works best when you can optimize your hardware
> so that the minimum amount of cycles are wasted, and this is easier when
> your workload is high enough to require most of the resources of one or
> more servers, most of the time.
>
> On a home server, you are likely to have limited requirements and lots
> of unused cycles, so minimizing total energy per device may be the right
> strategy, expecially when only one low-powered server-ish device is
> basically enought for the needs of the whole home and having more
> powerful device would just mean more wasted power.
>
> Of course, the economy of scale works in favour of power consumptions
> of big datacenters, but for certain kinds of data they are just not a
> valid alternative.

Yeah, so the point is that massive data centers do not in fact waste
energy (unless we want to go into Jevons Paradox [1]), but we are
trading efficient computing for other values, like control of our own
data.

Or to embrace and extend the FSF's recent slogan: There is no Cloud,
there's only other people's (very convenient and efficient!)
computers.

[1] https://en.wikipedia.org/wiki/Jevons_paradox