Server Load – Balancing using Resources of Clustering Technique

Abstract

Load Balancing is one of the most important issues for clustered servers. Load Balancing is done on the basis of server traffic. The proposed algorithm uses user-session history as the other constraint which is very useful for resource management over the server and easy access for the client. Collection of the data of a particular client from the last user-session, groups the applications that are likely to be opened in a specific time frame. This will provide easy access to the client and will reduce resource wastage on the server by using cluster technology. The primary clusters will be based on the bandwidth of the client’s internet bandwidth, which will map to different servers. Whereas secondary clusters will be based on the similarity of application-usage in the past user-session. This approach can reduce the unnecessary searching time. Cluster management will provide better approachability towards the server also the bandwidth based load balancing will lead to the minimum bandwidth wastage. Keywords Clustering, Load Balancing, Scalability, Client-Server, Resource Management, Cluster Creation, Cluster allocation, Cluster Reallocation, Cluster Management and Cloud Computing 1. INTRODUCTION The inspiration behind any Client-Server technology, arguably is its capacity to handle n no. of transactions in a given span of Time. Today, in the much credited dynamic environment “Scalability” is the concern for modern day Computer Scientist and Algorithm Pundits. You have only few ways left to deal with performance issues. 1. Either you are forced to upgrade hardware resources or 2. A load balancing [1] algorithm to find smart costefficient ways to fetch the user their desired resources without compromising Quality of Service. Coming directly to Load balancing on the Server, various methods have been used in past to balance the same, yet some haven’t got desired outcome. Clustering technique had been used in the near past for better fetch and efficient server loadmanagement Clustering Technique [2] is exploratory technique used in many areas to analyze large data sets. Given a proper notion of similarity, they find groups of similar variables/objects by partitioning the data set in “similar subsets”. Typically, several metrics over which a distance measure can be defined are associated with points (named samples) in the data set. There usually are various Clustering techniques available. Most used amongst them are I) Hierarchical and II) Partitioning clustering techniques. User Session Deduction is one of most important criteria for Load-Balancing, given the technique to be used is Server-Clusters. User-Session Identification has been non-trivial task to be performed on-the-go real time. Applications such as telnet or ssh typically generate a single TCP connection per single user-session, whereas application layer protocols such as HTTP, IMAP/SMTP and X11 usually generate multiple TCP connections per user-session. User-session characterization genuinely allows researchers to build realistic scenarios when assessing the performance of a complex network via simulation. Earlier Problems included I) Missed user Sessions, as threshold value had to be statically stated to tell the server that it is running over capacity. Vague, as it may sound, any usersession could be missed once there is bottleneck of connections asking permission for a connection establishment. II) Possibility of a huge data lose once there is a system failure since no backup image available, only to add to the chaos, a main server did not have any backup server available to route traffic towards that server. III) Every returning User had to start-over every time he/she terminates the session, it led to huge delay in page requested and page Fetched. As every user was forced to visit the prespecified pages and not the user-desired page. Our Approach will ensure a proper user session deduction takes place. The Algorithms will make sure to time stamp every usersession so that in case of any missing sessions the timestamp will auto-correct it’s reading. Apart from these a smart Artificial Intelligence Engine will predict user behavior with respect to their usage at different point of times. This will help user to directly visit his/ her required page rather than logging in again and getting diverted once again to Homepage. Also a smart cluster allocation Engine and cluster management algorithm will work hand in hand and deliver realtime cluster allocation to users without any hassles whatsoever. Even Cluster Allocation depends upon various parameters viz. Bandwidth, Predicted userbehavior. International Journal of Computer Applications (0975 – 8887) Volume 125 – No.5, September 2015 26 Fig 1: Architecture Diagram