Distributed
object-store, when you write an object it gets replicated across several
storage nodes (server) and across at least two data centers. The reason we do this is to meet
availability and durability modes. If
hard drive crashes data is on several servers so you can still access it, a
server can go down, a data center could do down, or a network split and S3 is
designed to perform despite these failures.
If a hard drive crashes S3 detects it and knows what objects were
replicated on that drive and instantly makes a new replica somewhere else in
the system. |
|
Of course this all
happens under the covers and as an application developer you don’t need to
worry about these failures. From your
perspective you see a highly available and durable data store. |
|
API is very
simple. Get, Put, Delete object…. |
|
Most folks take
anywhere from a few hours to a few days to integrate with S3—either complex
existing internet sites all the way to brand new startups. |