Re: Excuse me...
If your hypothetical streaming service only supports one person then you are correct. But they don't - they support hundreds of thousands of concurrent streams, pretty much 24*7.
The comparison is therefore with *all* of those local boxed being powered up (periodically or permanently).
If everyone had a USB stick in the side of their TV with the media on it then you'd probably draw ahead - but you haven't accounted for the resources associated with getting that data to the USB stick in the first place. If people have a NAS of some sort then you start to lose, badly.
"one" disk storing a video is going to be more power efficient than thousands of disks doing the same.... Yes I know there will be dozens of copies on Netflix servers, but the number will still be orders of magnitude lower than the number of people streaming each video.
The cost of running network switching gear isn't actually that high. A very brief squint at Cisco's product pages* suggests that 1W per Gbps is going to cover three hops... So that's 20 HD streams over 3 hops for 1W.
Netflix do their level best to minimise the hops, but a typical disk (Assuming spinning rust for a personal video storage solution) takes ~5W.
That's 300 "stream hops" in terms of network cost, and assumes you only fire up one disk. It also ignores any other components in your server, which of course are far better distributed at scale, further reducing your efficiency.
Of course Netflix uses disks as well - but they will be using them far more effectively than a home user.
The comparison isn't necessarily easy, but I certainly don't think it comes down easily on the side of local storage *for this application*.
*
> The typical power consumption per 10-Gigabit Ethernet port is less than 3.5 watts (W).
> The typical power consumption per 40- and 100-Gigabit Ethernet port is less than 14W and 22W respectively.
Biting the hand that feeds IT
