Decentralized Internet Considered Harmful

Today I was reading a really great story on TechCrunch about how the future of the internet could be serverless. I was immediately hooked in by the idea, myself being a big fan of P2P technologies that eliminate server usage. I've often pondered how this type of network could be built in the past, mostly coming up blank, but was excited that a company called MaidSafe might have figured it out.

MaidSafe is a fully decentralized platform on which application developers can build decentralized applications. The network is made up by individual users who contribute storage, computing power and bandwidth to form a world-wide autonomous system.

Their solution is essentially to replace server storage with a P2P-like network, where chunks of data are stored over the computing devices of many individuals. They go on to make promises about the inherent security of such a network, how it would provide an alternative revenue stream for developers, yada yada yada.
Though an attractive idea, I immediately began questioning its feasibility. Two points in particular make it seem unfeasible to me: the storage capacity  requirement from an end-user's machine, and the speed such a network could possibly have.
So I started to do the math. There is, as of March 2012, an estimate of about 50M servers active worldwide. The average server today has a storage capacity anywhere from 6TB to 10TB. Let's say 6TB, for our calculation. Total, that's about 300 exabytes of storage. Additionally, there are approximately 2.4B internet users. Let's say they all switch over to MaidSafe. So we get to distribute those 300 exabytes among 2.4B users, which yields approximately 125GB of storage per user. Now, there's three important caveats to this calculation I must clarify:

• The MaidSafe network replicates each piece of storage 4 times for redundancy. Initially I thought that meant the 300 exabytes of storage would have to be replicated 4 times, but that would be ignoring the fact that redundancies already exist in our server-based network, so I decided to exclude that bit of the calculation to be safe. Of course, not every piece of data is currently replicated 4 times, so you can imagine that the 125GB of storage required per user is on the low-end. 
• Additionally, the fewer people on this network, the higher that storage requirement to function properly. 
• Finally, though there's 2.4B active internet users, there's about 10B internet-connected devices. This means the storage requirements per-device would actually be closer to 30GB. That means your laptop, smartphone, and tablet would combined have to set aside 90GB of storage space. Though storage capacities on devices grew rapidly for a while, the switch to solid state drives is actually shrinking the average amount of storage capacity on a laptop computer due to expense, and I'm not convinced the average user would want to give up that kind of space. 

My second point is much more brief, I promise: the speed of a network needing to pull information from many disparate machines would be ridiculous. Not only would it require you to pull data from potentially hundreds of different machines per request, you would be subject to the speed of the other user's internet connection and the read speed off of their hard drive. Meanwhile, the servers we use today are optimized for fast reads, and generally you're pulling information off of one single server (ignoring any external requests the page you download might make).
While much of the data I've used here relies on estimates, it is simply to illustrate how unfeasible such a network would be. I'm nonetheless very excited by MaidSafe and where they might be headed. I encourage you to read the entire TechCrunch article if you get a chance, as there's many cool aspects of MaidSafe I didn't talk about. The idea of a serverless future sounds amazing, but I'm not convinced we're quite there just yet.

If I missed anything, please let me know! I look forward to your responses or corrections.