Software and scalability: It is all a numbers game.

A fascinating talk by Martin Thompson on making scalable software:

It's all a numbers game -- the dirty little secret of scalable systems by Martin Thompson

My highlights from the talk...

At a high level:

• Be simple
• Share Nothing Architecture
• Profile the code once a week.
• Separate reads from writes
• Know the platform
• Use cache oblivious algorithms

From a pattern perspective:

• Use the "Single Writer Principle": Any item of data, or resource, is only mutated by a single writer/thread.  It is OK if multiple threads, or other execution contexts, read the same data. CPUs can broadcast read only copies of data to other cores via the cache coherency sub-system. This has a cost but it scales very well.
• Disruptor vs queues: See the LMAX project on GitHub (http://lmax-exchange.github.io/disruptor/). Associated technical paper http://disruptor.googlecode.com/files/Disruptor-1.0.pdf.
• The "Curse of Logging": Most common Java Logging libraries take way too many CPU cycles from the application (Log4j, JUL, Logback, etc.). 
• Command Query Responsibility Segregation (CQRS):  Split the conceptual model into separate models for update and display.  This refers to as Command and Query respectively following the vocabulary of CommandQuerySeparation. 

Read Martin's blog.