Clarification Request: Performance Benchmark Comparing to "AWS S3 Files"

[norpol]

Thank you for this innovative project, I've been postponing giving it a shot for a while - but it definitively sparks a lot of creative use case ideas!

It appears this wasn't unnoticed by AWS engineers and they've kinda implemented their own (partial) ZeroFS alternative.

It's called AWS S3 Files a native FS built on top of S3, to quote their pricing page:

As you work with specific files and directories through the file system, associated file metadata and contents are placed onto the file system's high-performance storage, in particular the portions that benefit from low-latency access. Many read operations bypass the file system entirely, with data served directly from your S3 bucket at standard S3 GET request rates with no S3 Files data charges. Your authoritative data always remains in your S3 bucket. When you write data, it is stored on the file system's highly durable high-performance storage and then synced back to your S3 bucket, keeping the file system and your S3 bucket consistent in both directions.

at 60$ per TB, writes can be expensive - although retrieval and long term cost might offset that. Compared to EFS pricing of course is much more predictable, in particular for later reading.

---

High-performance storage* $0.30/GB-mo
File reads from high-performance storage $0.03/GB
File reads directly from S3 bucket** FREE
File writes $0.06/GB

[Barre]

S3 Files is much closer to "a synchronisation service between EFS and S3", it literally is EFS under the hood (same pricing for the hot tier), with S3 as the backing store. So it's not really a "filesystem on top of S3" the way ZeroFS is.

Compared to ZeroFS it'll always be more expensive (you pay EFS surcharge on top of S3 for any hot data), and it only runs on AWS compute. Performance-wise, I expect it to behave like EFS for cached data and much slower for data that is not cached. See https://www.zerofs.net/zerofs-vs-aws-efs

Also worth noting: writes to S3 Files don't get S3's durability guarantees immediately, they land on the EFS tier first and sync to S3 asynchronously (AWS says "within minutes"). So there's a window where your write exists only on EFS, and readers hitting the S3 API directly won't see it yet. ZeroFS writes go straight to S3, so you get S3's durability from the moment the write is acknowledged.