A lightweight interface to local grid scheduling systems

Abstract

Many complex research problems require an immense amount of computational power to solve. In order to solve such problems, the concept of the computational Grid was conceived. Although Grid technology is hailed as the next great enabling technology in Computer Science, the last being the inception of the World Wide Web, some concerns have to be addressed if this technology is going to be successful. The main difference between the Web and the Grid 111 terms of adoption is usability. The Web was designed with both functionality and end-users in mind. whereas the Grid has been designed solely with functionality in mind. Although large Grid installations are operational around the globe, their use is restricted to those who have an in-depth knowledge of its complex architecture and functionality. Such technology is therefore out of reach for the very scientists who need these resources because of its sheer complexity. The Grid is likely to succeed as a tool for some large-scale problem solving as there is no alternative on a similar scale. However, in order to integrate such systems into our daily lives, just as the Web has been, such systems need to be accessible to "novice" users. Without such accessibility, the use and growth of such systems will remain constrained. This dissertation details one possible way of making the Grid more accessible, by providing high-level access to the scheduling systems on which Grids rely. Since "the Grid" is a mechanism of transferring control of user submitted jobs to third-party scheduling systems, high-level access to the schedulers themselves was deemed to be a natural place to begin usability enhancing efforts. In order to design a highly usable and intuitive interface to a Grid scheduling system, a series of interviews with scientists were conducted in order to gain insight into the way in which supercomputing systems are utilised. Once this data was gathered, a paper-based prototype system was developed. This prototype was then evaluated by a group of test subjects who set out to criticise the interface and make suggestions as to where it could be improved. Based on this new data, the final prototype was developed firstly on paper and then implemented in software. The implementation makes use of lightweight Web 2.0 technologies. Designing lightweight software allows one to make use of the dynamic properties of Web technologies and thereby create more usable interfaces that are also visually appealing. Finally. the system was once again evaluated by another group of test subjects. In addition to user evaluations. performance experiments and real-world case studies were carried out on the interface. This research concluded that a dynamic Web 2.0-inspired interface appeals to a large group of users and allows for greater flexibility in the way in which data, in this case technical data, is presented. In terms of usability-the focal point of this research-it was found that it is possible to build an interface to a Grid scheduling system that can be used by users with no technical Grid knowledge. This is a significant outcome, as users were able to submit jobs to a Grid without fully comprehending the complexities involved with such actions, yet understanding the task they were required to perform. Finally, it was found that the use of a lightweight approach in terms of bandwidth usage and response time is superior to the traditional HTML-only approach. In this particular implementation of the interface, the benefits of using a lightweight approach are realised approximately halfway through a typical Grid job submission cycle.