S3 is locally redundant; buckets live in a specific region you put them in. You could roll your own geographically redundant storage by mirroring the same data into buckets in two or more regions, though. With the new pricing, two-region mirroring would run you between $0.136/GB/mo and $0.17/GB/mo, depending on whether you also wanted local redundancy within each region or were using "reduced redundancy storage" for each copy.
Joyent's pricing is slightly better if you want to roll your own multi-region storage: http://www.joyent.com/products/manta/pricing (Also comes with an interesting Unix-compute service where you can submit jobs to run over the data where it lives, rather than having to download it into a VPS to process, which I find more interesting than the storage itself.)
US Standard S3 region is geographically redundant. It routes to both Northern Virginia and "facilities in the Pacific Northwest." This redundancy is why it does not provide read-after-write consistency like the other regions.
That's what I thought as well, but recently found out otherwise. While it is true that, for buckets in the US Standard region, requests are routed to the closest endpoint, the data still only lives in one location.
To clarify by way of an example, say you're putting data into S3 from an EC2 instance in us-east-1. Those requests will end up being handled by the NVA US Standard S3 endpoint, and the data will only be stored in NVA. If try and retrieve that key from an EC2 instance in us-west-2, that request will indeed be routed to the PNW S3 endpoint, which will see that it doesn't have a local copy of the object and will need to retrieve it from NVA before serving it to you. That object will be cached locally in the PNW endpoint for an unknown amount of time, after which it will need to be retrieved from NVA again.
All of this information was from a recent support ticket I had open with AWS while trying to troubleshoot poor S3 performance.
The verbage they use to describe the US Standard region is quite confusing, leading a lot of people to assume that it provides geographic redundancy when it actually does not provide this at all.
Ah interesting, thanks! I didn't see that meta-region mentioned on the Pricing page (http://aws.amazon.com/s3/pricing/). Is it priced the same as the two lower-cost U.S. regions (N. Virginia / Oregon) that it apparently overlays?
Oregon (us-west-2) is a separate region. I don't know if the US Standard "Pacific Northwest" location actually is in Oregon, but even if it is, it's classified separately.
us-east-1 is the S3 "US Standard" meta-region. For all other services, us-east-1 is just good old "US East (N. Virginia)". For S3, the two names refer to the same thing. It uses the us-east-1 pricing, no matter where your data is physically located, since it is us-east-1.
Joyent's pricing is slightly better if you want to roll your own multi-region storage: http://www.joyent.com/products/manta/pricing (Also comes with an interesting Unix-compute service where you can submit jobs to run over the data where it lives, rather than having to download it into a VPS to process, which I find more interesting than the storage itself.)