The most suitable implementation depends on the desired properties of a queue. Should it be unbounded or a bounded one is fine? Should it be linearizable, or less strict requirements would be fine? How strong FIFO guarantees you need? Are you willing to pay the cost of reverting the list by the consumer (there exists a very simple implementation where the consumer grabs the tail of a single-linked list, thus getting at once all items put by producers till the moment)? Should it guarantee that no thread is ever blocked, or tiny chances to get some thread blocked are ok? And etc.
Some useful links:
Is multiple-producer, single-consumer possible in a lockfree setting?
http://www.1024cores.net/home/lock-free-algorithms/queues
http://www.1024cores.net/home/lock-free-algorithms/queues/intrusive-mpsc-node-based-queue
https://groups.google.com/group/comp.programming.threads/browse_frm/thread/33f79c75146582f3
Hope that helps.
