Ironically, it's possible to get a huge gain over purely random load balancing by examining just two queues at random -- essentially, you should always be doing

Ironically, it's possible to get a huge gain over purely random load balancing by examining just two queues at random -- essentially, you should always be doing this since the cost is O(1) and the improvement is large.[0] This doesn't require any distributed locking and at least would qualify as "intelligent" routing -- probably the bare minimum needed to justify that marketing label. Oh, and it also scales incredibly well. Like I said, there's no reason not to use it over purely random load balancing. [0] http://www.eecs.harvard.edu/~michaelm/postscripts/mythesis.pdf http://www.eecs.harvard.edu/~michaelm/postscripts/mythesis.p...