I’ve been asked in my lecture on distributed systems “you keep telling us that distributed systems are harder, so why go for a peer-to-peer approach if we can just rent a server?”.

This year I let the students calculate the cost of different solutions. The result: If you don’t have an average payoff per function invocation, cloud services are a really bad deal. Why did many big changes originate in peer-to-peer? Because that’s where you can actually scale up without having to monetize to the brink. And if your function runs amok, your users just restart your program and some of them might even report a bug.

Expand full comment

Sorry Forrest, but I would call this a cowboy approach.

As engineering director in several startups in the past I would say that the risk of putting our company to the brink of bankrupcy is too high. This is just irresponsible.

If there is no way to limit spending then this tehcnology is too immature.

On Azure were was a Dedicated (App Service) Plan, so I could have a hard limit of what I could spend.

The "infinite" scalability is rerelly needed. I need a decent perfomance and I'm ok with service degratation on slashdot effect.

Business values predictability much more than opportunity that rarelly can be used.

Expand full comment