Amazon S3 - Design

Design

Details of S3's design are not made public by Amazon. According to Amazon, S3's design aims to provide scalability, high availability, and low latency at commodity costs.

S3 is designed to provide 99.999999999% durability and 99.99% availability of objects over a given year., though there is no SLA for durability.

S3 stores arbitrary objects (computer files) up to 5 terabytes in size, each accompanied by up to 2 kilobytes of metadata. Objects are organized into buckets (each owned by an Amazon Web Services or AWS account), and identified within each bucket by a unique, user-assigned key. Amazon Machine Images (AMIs) which are modified in the Elastic Compute Cloud (EC2) can be exported to S3 as bundles.

Buckets and objects can be created, listed, and retrieved using either a REST-style HTTP interface or a SOAP interface. Additionally, objects can be downloaded using the HTTP GET interface and the BitTorrent protocol.

Requests are authorized using an access control list associated with each bucket and object.

Bucket names and keys are chosen so that objects are addressable using HTTP URLs:

• http://s3.amazonaws.com/bucket/key
• http://bucket.s3.amazonaws.com/key
• http://bucket/key (where bucket is a DNS CNAME record pointing to bucket.s3.amazonaws.com)

Because objects are accessible by unmodified HTTP clients, S3 can be used to replace significant existing (static) web hosting infrastructure. The Amazon AWS Authentication mechanism allows the bucket owner to create an authenticated URL with time-bounded validity. That is, someone can construct a URL that can be handed off to a third-party for access for a period such as the next 30 minutes, or the next 24 hours.

Every item in a bucket can also be served up as a BitTorrent feed. The S3 store can act as a seed host for a torrent and any BitTorrent client can retrieve the file. This drastically reduces the bandwidth costs for the download of popular objects. While the use of BitTorrent does reduce bandwidth, AWS does not provide native bandwidth limiting and as such users have no access to automated cost control. This can lead to users on the 'free-tier' S3 or small hobby users amassing dramatic bills. AWS representatives have previously stated that such a feature was on the design table from 2006 to 2010 but have recently stated the feature is no longer in development.

A bucket can be configured to save HTTP log information to a sibling bucket; this can be used in later data mining operations. This feature is currently still in beta.