Ahead in the Cloud by Werner Vogels
Ahead in the Cloud – The power of Infrastructure as a Service
CTO Amazon.com, Dr. Werner Vogels
Pretty much everyone in the audience uses Amazon!
Announced: Persistent Storage for Amazon EC2.
Hitting one page, might actually go to 250 different services, before the page is generated for you. Shows the use of a tool (Amazon internal), that graphs it.
SaaS: Develop -> Test -> Operate
Hardware costs? Software costs? Maintenance? Load balancing? Scaling? Utilisation? Idle machines? Bandwidth management? Server hosting? Storage management? High availability? All this is the differentiated heavy lifting that Amazon bases their services on.
SaaS comes at a very big cost that you have to address.
70/30 switch: 30% of time, energy and dollars on differentiated value creation; 70% of time, energy and dollars on differentiated heavy lifting.
At Amazon, we expect data centres to fail. But we also expect software to tolerate this failure.
“Scalable Infrastructure that allows applications to meet infinite demand, cheaply and reliably” (statement, made with picture of large amount of Sun hardware)
Amazon S3 (storage), SimpleDB, EC2 (computer power), FPS (payment service). All this is scalable (increase/decrease capacity on demand).
Scalability. Availability. Performance. Cost-Effectiveness.
Growth: largest selection on earth, good customer experience, drives prices down, drives traffic, sellers, selection, and this is a cycle for growth. It brings a lower cost structure, that also lowers prices down then.
This means that incremental scalability is key to Amazon’s business. Grow one step at a time, consistently. Turn a fixed cost, into a variable cost, as your business grows seamlessly.
Elastic cloud: grow and shrink on demand, with minimal disruption to performance. Operational efficiency, fault-tolerant, and remember, everyone has different growth paths. Heterogeneity – do not believe that all your nodes have the same capacity! A year from now, you will have more powerful machines, your software must scale for this.
Everything fails, all the time. An epic truth.
Failures are highly correlated. By every possible worst way! Systems do not fail by stopping – they will fail by sending out garbage ;) Your system must be able to deal with that.
Determinism is an illusion. An illusion created in a very small space. “Let go of control!”
Engineer for performance at 99.9%. Remember, address uncertainty – acquire resources on demand, pay for what you use, leverage other’s core competencies, turn fixed costs into variable costs. Never every pay again for something sitting in your data centre doing nothing for you.
All data access at Amazon is primary key based. Eventual consistent, for high read volume and always writeable. Query-based access, was non-relational.
Primary Key Access: Amazon S3; Query-based Access: SimpleDB; EC2 with persistent storage for a dedicated solution
Persistent storage? Raw disk, attach a volume to EC2. You can also detach. Infinite scalability in terms of data. From snapshots, you can create new volumes.
“All you need is a credit card” – for AWS. Lots of laughter :)