Weird article, or maybe just a weird title. This author talks about how hard it is to support locking in complex datastructures built on top of DHTs. That doesn't seem like either what DHTs are good for or what a "killer app" needs. Use a DHT for implementing your tinyurl clone, not for inventing a prefix-range-lookup database.
Not only that, the paper he references used OpenDHT as its substrate. OpenDHT was Sean Rhea's research project and was never designed for performance or for the particular operations the paper authors tried to use it for (specifically, the data structure they were attempting to implement required an atomic write assurance for key insertion and they were forced to add a locking layer on top of the DHT which was basically a set-and-test for a lot of the put operations...) This is much akin to those people who try to store unstructured data in a RDBMS and then bitch and moan about how all RDBMSs must suck because of the poor performance they are seeing.
I know the blog author has a particular agenda to push, but one would hope that the least we could assume is that he actually read and understood the details of the paper he uses to bolster his claims...
