Recorded at QCon last year, this presentation by Martin Thompson starts with an intro to caches on modern microprocessors, then talks about Consistency models, the x86 memory model, and then talks about languages Memory Model (Java, C++11, Erlang).
The presentation then passes over to Michael Barker which talks about Model Specific Registers.
Martin then talks about Contention and Amdahl's Law, locks, atomic instructions. He then goes into what he calls "lock-free algorithms", but he only covers Queues. He does a comparison study of performance an latency for the ConcurrentLinkedQueue, ArrayBlockingQueue, LinkedBlockingQueue, and others.
Michael then takes back control, and talks about False Sharing and some performance benchmarks they did with and without padding.
http://www.infoq.com/presentations/Lock-free-Algorithms
If you don't know who Martin Thompson is, then check out his blog:
http://mechanical-sympathy.blogspot.co.uk/
and Michael Barker's http://bad-concurrency.blogspot.co.uk/