Delete comment from: DSHR's Blog
David,
I think you're ignoring two major difference between storage and minding: latency and bandwidth.
In mining, blocks are small, and can be easily transferred over long distances between pools very quickly. This means mining in remote areas of the globe has no drawback compared to mining close to cities. So mining concentrates in areas where it is most efficient, causing centralization of hardware. (Mining pool centralization is a separate but related issue).
Your post makes a similar argument for storage centralization, where storage will centralize where it is more efficient. However, location matters for storage.
Imagine that there was a data center in China that could offer unlimited storage 100x more efficiently then anywhere else on the globe. You might naively assume then that they would have ~100% of the global storage market. However, if I wanted to store a movie in that center and stream it to my laptop in New York, I would run into unsolvable latency and bandwidth problems. Having everyone on the planet use this center only makes the problem worse, as the bottleneck between the US and China are the international undersea cable relays, which would be permanently clogged trying to service the whole planet. Ultimately, such a data center would end up as a giant tape drive storage center. Not useless, but not for universal use.
A local entity is always going to be able to offer better latency and bandwidth costs then a distant entity, just by basic physics. Since all data centers use commodity hardware from many competing low-margin manufacturers, close access to equipment manufacturing is not a large advantage. In addition, electricity costs are much lower % of the cost of providing storage as opposed to mining, so I do not think having access to cheap electricity is going to be as large a barrier to energy as with mining.
I do think there are economies of scale that can be accessed. Using dedicated server racks, a dedicated internet connection, and ~99% uptime are probably going to be required. I do not think "random desktop drives" is a reasonable economic model. But I also think small data centers, operating through a network that takes care of overhead, so data becomes a pure commodity like gasoline, have a strong economic case.
AWS doesn't already do this because the marginal overhead cost of setting up a data center in every city is high, and the cost of making a mistake and damaging their network is large (as has already happened). But if the protocol could handle nodes entering and leaving the network with reasonably strong stability and redundancy guarantees (ideally, it would be plug and play), then there is no overhead for starting up your own center to compete in your local network, where you can offer competitive bandwidth and latency rates.
Sorry for the long rant. I'm not an expert in the field, so any of my assumptions here might be wrong, but this is the case I have been thinking of for some time.
Jun 21, 2018, 11:31:34 PM
Posted to The Four Most Expensive Words in the English Language