Thanks for sharing your writeup and the potential use cases therein.

We've been building an experimental shared file system that specifically targets the FaaS setting. It supports SQLite and gets big performance gains over NFS/EFS, especially on read-mostly workloads, due to improved local caching and lock elision. See: https://arxiv.org/abs/2009.09845

Really interesting.

How do you deal with ephemeral file storage? I guess you actually need the provider (say AWS) to build this into the platform?

Yes, the idea is that the cloud provider should offer something like this. We intercepted system calls to simulate what it would look like.

Did I read it right you get 1 second responses? Why even bother with lambdas? A t3 nano reserved instance costs 2 bucks a month or something! Surely most folks don't anticipate 1000x scaling in minutes? Coupled to elastic beanstalk you can get reaonable scaling as well!

Might be useful for services that aren't continually used? For example, month or year end processes. Not convinced I would use SQLite for a service like that though, seeing as AWS has serverless RDS

This was just a proof-of-concept using the smallest lambdas available. I definitely would opt for a EC2 instance. Though lambda would be enough for something like a configuration service.

Is a lambda easier to configure and set up? With Elastic Beanstalk and flask (and the sample code they provide) I can get an api running in a few minutes..

Both are trivial and take only minutes if you're already familiar with them, but Lambda doesn't involve managing instances, their OSes, or making sure they scale up / down (at the instance level) the way you want.