I think you just might have a legit candidate for a MapReduce-like algorithm here since your work seems to be a great candidate for parallel work. Check the Had

I think you just might have a legit candidate for a MapReduce-like algorithm here since your work seems to be a great candidate for parallel work. Check the Hadoop article that was posted here not too long ago. It will let you scale the work over many nodes in a fairly linear fashion. You might not even have to go that route if you properly shard your data among your computers. If you store user profiles based on locality or the like, the work is guaranteed to be local. Also, you can try segmenting your users based on general factors that group them (a course version of your matching algorithm) to preserve locality.