New Replica Server Placement Strategies using Clustering Algorithms and SOM Neural Network in CDNs

Ghazaleh Eslami1*, Abolfazl Toroghi Haghighat1, Saeed Farokhi2

1Department of Computer Engineering, Islamic Azad University, Iran

2Department of Mechanical Engineering, Tehran University, Iran

Abstract: Many service providers distribute various kinds of content over the internet. Content Distribution Networks (CDNs) use replication of either entire website or most used objects to bring content close to the users and improve communication delay.  In order to deliver web contents, CDNs should decide where to place replica servers and how many replicas are needed.  In this paper, a linear programming formulation for web server replica placement has been provided.  We also present new algorithms using K-means, Fuzzy c-means clustering and SOM Neural network to place web server replicas. Our objective is to find best replica server sites, which minimize distance between replicas and clients- to keep replicas. We compare our algorithms with Greedy algorithm. We have considerable enhancement in terms of load balancing and Runtime.  

 Keywords: Distributed systems, server placement, clustering algorithms, CDN.

 Received January 6, 2014; accepted June 10, 2015