Distributed Hash Table (DHT) systems are scalable and efficient data structures for object storage and location using a simple put/get interface. These systems place objects over a very large set of hosts using a multitude of algorithms in order to distribute objects uniformly among hosts using logarithmic (or lower) costs for routing table sizes and message hops [1, 2]. However, these systems assume that object size (storage load) and popularity (communication load) follow an uniform distribution. When unbalanced data is used on a DHT, hotspots are created at some specific (random) hosts. Although one might argue that storage is not a critical resource, due to the current trend on secondary storage capacity, storing such large objects creates network bottlenecks, which in turn may limit data availability.
10.1.1.163.9683