MOSIX: Implementation, Trend and Benchmark in Malaysia M. Hakim, J. Jais, S. Salwa College Of Infonnation Technology Universiti Tenaga Nasional Email: mohdhakim@uniten.edu.my.iamilin@uniten.edu.my.sitisalwa@uniten.edu.my Abstract This paper is part of a larger project on MOSIX cluster computing. This paper also is a continuation of the previously published paper regarding MOSIX potential performance and threat [9]. The Inain focuses are on MOSIX implementation, survey and benchmark on MOSIX performance. In this paper, the information is presented in a broad to narrow approach, which is froln an overview of shared network and all the way to the MOSIX. We also test several testing algorithm to benchmark the perfonnance of MOSIX in order to find out the real potential of MOSIX distributed operating system. Keywords: MOSIX, Distributed Computing, Parallel Computing, Clustering 1. Introduction Parallel processing has been around for quite awhile and there were many books by respected authors that helped in getting a good grounding of the history, the evolution and operation. While looking for relevant Inaterial on MOSIX, all the infonnation gathered was from the World Wide Web. The MOSIX homepage at http://www.mosix.org was especially helpful for general information and for technical details. Usually, a distributed computing architecture consists of lightweight software agents installed on client systems (for example: PCs), and dedicated distributed computing management servers. The distributed cOInputing works when the agent running on the client system detects that the client is idle. It notifies the management server that the client is free for processing, and requests an application package. The client then receives the application package from the server, runs the software when it has spare CPU cycles, and sends the results back to the server. The application runs in the background of the client system and does not have an effect on normal use of the computer. If the user of the client system needs to run 978-1-4244-2328-6/08/$25.0002008 IEEE any of his own applications, control is immediately returned, and processing of the distributed application package ends [2]. An important fact to know is that not all resources of a distributed systeln are devoted to solving a single problem. A significant fact that distinguishes a distributed system fro In parallel processing is that it is wide ranging and can work on many problelns simultaneously. All the computers connected within this systeln are able to function independently, but can share their data. The most popular application of distributed cOInputing is SETI@HoIne; which analyzes data fro In a telescope to search for potential signs of extraterrestrial life. In the commercial world, distributed computing is called client/ server computing. The overview is now slightly narrowed down to parallel computing in the section below. 2. Clustering Clustering is a sub-paradigm of parallel processing and distributed cOInputing. Networked computers, on a local level, work together just like a parallel machine. Although the work is split across multiple computers, they behave in a parallel fashion. To put it more clearly, a definition of a cluster is a type of parallel or distributed systeln that: a) Consists of a collection of interconnected computers, b) Is used as a single, unified computing resource [3]. For this project; a cluster will be developed that will utilize parallel processing. As also stated previously, parallel computing is not widely used because due to the difficulty of programIning for parallel machines. This is where the MOSIX Cluster Management System comes in.