Still, even if the patents were in some way novel at their time, I think the details are far more important, especially: how close is AWS' implementation to the exact algorithm they describe? How many others came to the exact same implementation not knowing about this patent, potentially proving that it didn't actually meet the bar for novelty at the time it was granted?

Still, patents become public from the moment they are filed. So you can scour the patent applications database as well if you fear others have already filed for a patent for something you are trying to do. Of course, that will mean that, if you go ahead with your own implementation and the patent is later granted, you will not be able to claim that you didn't know about this, and will face higher damages if you end up in a lawsuit.

The principle here is that a company decides to publish some internal discovery instead of keeping it for themselves, and the government in return grants them a limited time monopoly on that invention. The company could instead keep the invention a secret, but in that case, they can't stop others from using it, and if someone else patents it, their secret invention can actually infringe on the patent granted to someone else later.

No. There is an 18 month delay to publication: https://www.uspto.gov/web/offices/pac/mpep/s1120.html

How do you do that? beyond my pay grade.

If you want more definitive timestamping, there are services that will do that, like OpenTimestamps, an actually a decent use case for blockchain.

https://en.m.wikipedia.org/wiki/Lustre_(file_system)

Also, depending on the feature set listed in the patent, GPFS (which I remember using around 2000) was definitely around earlier than 2000. It was/is a clustered file system from IBM that had distributed storage nodes. Not sure how it did metadata though, which might not make it relevant to the patent, but the same ideas were definitely at work here.

(And it being IBM, you can be sure those were patented too)

https://en.m.wikipedia.org/wiki/GPFS