The initial architecture of Amazon.com was a single monolith with a battery of relational databases that created a significant scaling bottleneck.
Amazon's first major architectural shift involved breaking the monolith into services, but an initial data-driven decomposition created new, large services that became bottlenecks themselves.
The move to smaller microservices improved developer productivity until the overhead of managing capacity and databases for each service led to the creation of internal infrastructure as a service platforms.
Systems must be designed with evolvability as a core principle, allowing for continuous change and feature additions without downtime, as exemplified by Amazon S3.
A cell-based architecture isolates failures by partitioning customers and infrastructure into independent, self-contained units, thereby limiting the blast radius of any single outage.
By mapping customers to multiple overlapping cells (sharding), the blast radius of a failure is reduced from a fraction of cells to a much smaller fraction based on mathematical combinations.
Frugality is a key architectural driver where cost is treated as a primary non-functional requirement, and making cost metrics visible to engineers changes their behavior and improves efficiency.
The choice of programming language has a significant impact on system cost and energy efficiency, with languages like Rust offering major advantages in performance and security over others like Python or Ruby.
22:24Florian Mair
30:35Werner Vogels, Mike Butcher & Daniel Gebler
26:22David Tielke
26:45Babette Wagner
26:34Billy Thompson
50:06Jens Happe
43:07Adam Mullen & John Collins
46:24David Leitner
.gif?w=240&auto=compress,format)
Jobs that call for the skills explored in this talk.
doinstruct Software GmbH
Berlin, Germany
Amazon.com, Inc
Amazon.com, Inc
Amazon.com, Inc
Berlin, Germany
Amazon.com, Inc