 |
Twenty years in the past right now, on March 14, 2006, Amazon Easy Storage Service (Amazon S3) quietly launched with a modest one-paragraph announcement on the What’s New web page:
Amazon S3 is storage for the Web. It’s designed to make web-scale computing simpler for builders. Amazon S3 gives a easy internet providers interface that can be utilized to retailer and retrieve any quantity of information, at any time, from wherever on the net. It offers any developer entry to the identical extremely scalable, dependable, quick, cheap information storage infrastructure that Amazon makes use of to run its personal international community of web pages.
Even Jeff Barr’s weblog publish was just a few paragraphs, written earlier than catching a aircraft to a developer occasion in California. No code examples. No demo. Very low fanfare. No person knew on the time that this launch would form our whole trade.
The early days: Constructing blocks that simply work
At its core, S3 launched two easy primitives: PUT to retailer an object and GET to retrieve it later. However the actual innovation was the philosophy behind it: create constructing blocks that deal with the undifferentiated heavy lifting, which freed builders to give attention to higher-level work.
From day one, S3 was guided by 5 fundamentals that stay unchanged right now.
Safety means your information is protected by default. Sturdiness is designed for 11 nines (99.999999999%), and we function S3 to be lossless. Availability is designed into each layer, with the belief that failure is at all times current and have to be dealt with. Efficiency is optimized to retailer nearly any quantity of information with out degradation. Elasticity means the system robotically grows and shrinks as you add and take away information, with no guide intervention required.
After we get these items proper, the service turns into so easy that the majority of you by no means have to consider how advanced these ideas are.
S3 right now: Scale past creativeness
All through 20 years, S3 has remained dedicated to its core fundamentals even because it’s grown to a scale that’s exhausting to grasp.
When S3 first launched, it provided roughly one petabyte of complete storage capability throughout about 400 storage nodes in 15 racks spanning three information facilities, with 15 Gbps of complete bandwidth. We designed the system to retailer tens of billions of objects, with a most object dimension of 5 GB. The preliminary worth was 15 cents per gigabyte.
Right now, S3 shops greater than 500 trillion objects and serves greater than 200 million requests per second globally throughout lots of of exabytes of information in 123 Availability Zones in 39 AWS Areas, for hundreds of thousands of shoppers. The most object dimension has grown from 5 GB to 50 TB, a ten,000 fold improve. In case you stacked all the tens of hundreds of thousands S3 exhausting drives on prime of one another, they’d attain the Worldwide Area Station and virtually again.
Whilst S3 has grown to assist this unbelievable scale, the value you pay has dropped. Right now, AWS prices barely over 2 cents per gigabyte. That’s a worth discount of roughly 85% since launch in 2006. In parallel, we’ve continued to introduce methods to additional optimize storage spend with storage tiers. For instance, our clients have collectively saved greater than $6 billion in storage prices by utilizing Amazon S3 Clever-Tiering as in comparison with Amazon S3 Normal.
Over the previous 20 years, the S3 API has been adopted and used as a reference level throughout the storage trade. A number of distributors now supply S3 appropriate storage instruments and techniques, implementing the identical API patterns and conventions. This implies expertise and instruments developed for S3 usually switch to different storage techniques, making the broader storage panorama extra accessible.
Regardless of all of this progress and trade adoption, maybe probably the most outstanding achievement is that this: the code you wrote for S3 in 2006 nonetheless works right now, unchanged. Your information went via 20 years of innovation and technical advances. We migrated the infrastructure via a number of generations of disks and storage techniques. All of the code to deal with a request has been rewritten. However the information you saved 20 years in the past remains to be accessible right now, and we’ve maintained full API backward compatibility. That’s our dedication to delivering a service that frequently “simply works.”
The engineering behind the size
What makes S3 doable at this scale? Steady innovation in engineering.
A lot of what follows is drawn from a dialog between Mai-Lan Tomsen Bukovec, VP of Information and Analytics at AWS, and Gergely Orosz of The Pragmatic Engineer. The in-depth interview goes additional into the technical particulars for individuals who need to go deeper. Within the following paragraphs, I share some examples:
On the coronary heart of S3 sturdiness is a system of microservices that constantly examine each single byte throughout the complete fleet. These auditor providers look at information and robotically set off restore techniques the second they detect indicators of degradation. S3 is designed to be lossless: the 11 nines design aim displays how the replication issue and re-replication fleet are sized, however the system is constructed in order that objects aren’t misplaced.
S3 engineers use formal strategies and automatic reasoning in manufacturing to mathematically show correctness. When engineers test in code to the index subsystem, automated proofs confirm that consistency hasn’t regressed. This identical method proves correctness in cross-Area replication or for entry insurance policies.
Over the previous 8 years, AWS has been progressively rewriting performance-critical code within the S3 request path in Rust. Blob motion and disk storage have been rewritten, and work is actively ongoing throughout different elements. Past uncooked efficiency, Rust’s kind system and reminiscence security ensures get rid of whole lessons of bugs at compile time. That is a necessary property when working at S3 scale and correctness necessities.
S3 is constructed on a design philosophy: “Scale is to your benefit.” Engineers design techniques in order that elevated scale improves attributes for all customers. The bigger S3 will get, the extra de-correlated workloads grow to be, which improves reliability for everybody.
Trying ahead
The imaginative and prescient for S3 extends past being a storage service to changing into the common basis for all information and AI workloads. Our imaginative and prescient is easy: you retailer any kind of information one time in S3, and you’re employed with it instantly, with out transferring information between specialised techniques. This method reduces prices, eliminates complexity, and removes the necessity for a number of copies of the identical information.
Listed below are a couple of standout launches from latest years:
- S3 Tables – Absolutely managed Apache Iceberg tables with automated upkeep that optimize question effectivity and scale back storage value over time.
- S3 Vectors – Native vector storage for semantic search and RAG, supporting as much as 2 billion vectors per index with sub-100ms question latency. In solely 5 months (July–December 2025), you created greater than 250,000 indices, ingested greater than 40 billion vectors, and carried out greater than 1 billion queries.
- S3 Metadata – Centralized metadata for fast information discovery, eradicating the necessity to recursively record massive buckets for cataloging and considerably lowering time-to-insight for giant information lakes.
Every of those capabilities operates at S3 value construction. You’ll be able to deal with a number of information varieties that historically required costly databases or specialised techniques however are actually economically possible at scale.
From 1 petabyte to lots of of exabytes. From 15 cents to 2 cents per gigabyte. From easy object storage to the muse for AI and analytics. By way of all of it, our 5 fundamentals–safety, sturdiness, availability, efficiency, and elasticity–stay unchanged, and your code from 2006 nonetheless works right now.
Right here’s to the subsequent 20 years of innovation on Amazon S3.