A Scalable Work Function Algorithm for the k-Server Problem

Sharath Raghvendra, Rachita Sowle

Abstract: We provide a novel implementation of the classical Work Function Algorithm (WFA) for the k-server problem. In our implementation, processing a request takes O(n²+k²) time per request; where n is the total number of requests and k is the total number of servers. All prior implementations take Ω(kn² +k³) time per request. Previous approaches process a request by solving a min-cost flow problem. Instead, we show that processing a request can be reduced to an execution of the Dijkstra’s shortest-path algorithm on a carefully computed weighted graph leading to the speed-up.