This is a great direction for a certain class of problems, but to call this a general approach for building a super computer ignores the biggest challenge of all: communication performance.<p>The leadership-class super computers have large budgets for network interconnects like Infiniband, because the hardest problems require ultra low-latency, high-throughput RDMA message passing between processes.
I actually made this a few years ago for an ill-fated application to YC (I was still a sophomore in college).<p>The first big problem is that you can't trust any of the results that you receive from the workers. You can alleviate this somewhat by not accepting job results until they've been replicated by at least several other workers (lowering efficiency of the system), but then you're still vulnerable to sybil attacks.<p>The second issue was the total lack of privacy for the data and computations.
Ok, this is similar to Gridcoin (<a href="https://gridcoin.io/" rel="nofollow">https://gridcoin.io/</a>) , Golem (<a href="https://cryptoslate.com/coins/golem/" rel="nofollow">https://cryptoslate.com/coins/golem/</a>) , SPARC (<a href="https://sparc.network/" rel="nofollow">https://sparc.network/</a>) , et al. I could keep going, but the point is, something like this exists. And we are seeing the issues of what happens when you embed a program that takes advantage of processing power. You get things like Coinhive (<a href="https://coinhive.com/" rel="nofollow">https://coinhive.com/</a>), which were meant to be benign, turned into botnets.
For those having trouble viewing the site: <a href="https://web.archive.org/web/20180303173013/http://ben.akrin.com/?p=5997" rel="nofollow">https://web.archive.org/web/20180303173013/http://ben.akrin....</a>
This will only work if you can cheaply validate the work performed. Or send it to several distinct visitors and compare results, driving down the likelyhood that the results are bogus.
Something like this could be a way to get finally rid of advertisements, and help finance free services in a novel new way if the CPU isn't over committed. I thought the same around crypto mining... Many people might prefer the resource providing/computation option compared to ads.
> Unlike a regular computer cluster, the nodes are very ephemeral (as website visitors come and go) and can’t talk to each other (no cross site requests).<p>Not sure how that would actually help but they somewhat _can_ talk to each other through WebRTC. I wonder if that would change anything here
This could be a good alternative for SETI or that protein folding project but if it somehow turns into a plague of attempts to shift the computing requirements of running the backend onto the users, our content blocking lists will have to grow a little bit longer again.
opportunity here to create cloud computation platforms where people submit work which gets executed on other people's machines who are getting paid to allow their browser to contribute compute ... eventually price of AWS / google compute engine will drop ... win win