[cjsaylor]

This is a cool idea. The price as listed is "cheap", but for the feature set, it's not really that cheap. It's half the cost of a digitalocean droplet with no security and possible surge pricing.

I think it's a fantastic idea and the implementation here seems easy to use, but needs to be significantly cheaper.

[grandrew]

cjsaylor, thank you for your feedback! Yes the price is not so low I just had to have some figures to start with.

But as you may know - the $5 machine you are referencing here is not suitable for any meaningful work as you will have about 300MB of RAM after you boot it for your use - and this is about the amount or RAM you will need just for your framework to load up. So you should basically compare it with the $10 or $20 machine. Also you should take into account that the price here is the actual CPU-time - while DigitalOcean charges you the wall-hour - so I estimate the price like 5x-10x lower depending on your task.

But yes I know where to move from here - for example switching to accounting in GigaFLOPS-hour rather than machine time to allow usage of power efficient laptops for example (old laptops are usually slower but are still very energy efficient) and also for the owners of faster hardware to benefit

The absence of any authentication on both sides is arguably a disadvantage though. You may choose to compute anonymously for privacy related reasons and Junk.Systems is the only platform that allows you to do so AFAIK.

[cjsaylor]

Fair points. I guess it wasn't super clear from the pricing example. I think, as you say, expressing the price per CPU time rather than wall clock time is what would better sell me on it.

I could actually see a pretty good use case for academic computing (think SETI at home, or the genome project) to where they could get cheaper access to CPU.