What are Cycle-Stealing Systems Good For? A Detailed Performance Model Case Study

The idea of stealing cycles has been hyped for some years, boasting unlimited potential by tapping the computational power of millions of under utilized PCs connected to the Internet. Despite a few spectacular success stories (e.g. SETI@HOME), cycle-stealing is today not a widely used technology. We believe two principal impediments need to be overcome. The first is ease of development and use. Most of the problems faced in developing cycle stealing applications are not specific to those applications, so generic cycle stealing frameworks such as our G2 framework can play a vital role in this regard. The second is uncertainty. Potential developers don't know whether if they went to the effort of developing a parallel application for a cycle stealing environment, it would pay off, i.e. whether they would get a reasonable speedup. To minimize this risk, we propose the development and use of detailed performance models.