It's interesting to see the continued trends to separate compute from storage.

I'm curious about the tradeoffs. The article mentions increased latency of 10's of milliseconds. If you have a complex access (many joins or index hits), is that exacerbated from many round trips to load different files? From a cost perspective, I image that you now have to think about IOPS. If you have a very high usage database, would it make sense to have a more traditional file system where you aren't paying per operation?

Edit: On reread, I missed that this is only available for the hypertable concept, and is more of a "cold storage" for older metrics, presumably accessed infrequently. I am curious about a general "PG backed by S3"'s ramifications.