Load Balancing Distributing File System Servers: A Rule-Based Approach

Abstract

In  this  paper,  we  present  a  new  methodology  for managing read-write filesets  across  multiple  file servers of a Distributed File System, thus balancing  the  load  of  file access requests across servers. The proposed methodology is based on a rule-based data mining technique and graph theory algorithms.   The  rule-based technique generates rules from access request data  to  identify  present file access patterns in the system. We then use the  rules,  graph  analysis and statistical information (usage and size of the  filesets) to relocate the filesets between different file servers. The algorithm for fileset relocation is based on the graph coloring problem. We tested our algorithms on data collected for five months on DFS file servers in  a  production  environment.   Experiments  with  the data show that our methodology  can  make intelligent decisions about file system transfers in order to balance the access request load across DFS servers.