BOINC on JXTA - suggestions for improvements

Distributed computing systems allow acquiring resources that lie outside the capabilities of even the largest supercomputers. Indisputable example is prosperity of SETI@Home project which gathered at its peak ca. 850,000 users. Its success is partially related to well designed architecture – BOINC. Many projects work on this architecture, using all together the computing power of ca. 400 TeraFLOPs. This is more than the most hi-tech supercomputer. But even BOINC architecture struggles with scalability concerns. The creators may not have predicted the degree of projects’ popularity. The purpose of this work is to find new solutions for volunteer computing. This paper contains overview of existing systems which elements might be useful for this goal Next it describes BOINC parts along with propositions of changes which can benefit its effectiveness. In the second part has been shown how the popular JXTA platform can be utilized to completely redesign the existing structure. A complete project of new architecture has been introduced which has the same functionality as BOINC, but it overcomes its limitations and the resources are utilized better. This work also involves partial implementation and points out the directions of future development. Part I – Existing solutions 1. Overview of selected systems for distributed computing and file sharing The goal of this chapter is show the solutions used in other systems that influenced the suggested improvements. Two systems for file sharing and three for distributed computing are presented. ● Napster – served as a P2P system for sharing MP3 files [18]. It was created in the year 1999 by Shawn Fanning in only few months. The idea was both simple and effective. A user that downloaded the application shared a directory with his MP3 files and was registered on Central Indexing Server. When a user was looking for a file, he contacted the server which in turn queried registered computers. If one of them had the needed file, it directly contacted the seeking computer and started the transfer. After the finish, it disconnected. Even though the server was used for indexing, the real file sharing was performed without its participation, so we call Napster a P2P system. Despite of small range of duties (seeking, not sharing data), the server became a bottleneck and limited scalability of the system. The possibilities and simplicity of use have started the big popularity of this program, which drawn the attention of producers of original albums with music. The activity of Napster was suspended for piracy reasons. Nevertheless it started a new era in resource sharing with the help of Internet. ● Kazaa – serves as a system for sharing files [22], created in the year 2001. Its potential is not limited to music files only: users can share any kind of data. Unlike Napster, Kazaa uses seek algorithm called FastTrack, which is called a 2 generation P2P algorithm. It introduces Supernodes, fast computers with big computational power and broadband connection to the net. A simple node connecting to the net registers at the Central Server and receives a list of active Supernodes. If it fulfils the requirements, it may be chosen as a Supernode itself, without notifying the user. When looking for a file (figure 1.1), a node contacts not the Central Server, but one of the Supernodes which in turn forwards the query to other Supernodes and simple nodes (a single Supernode can serve 60 to 150 simple