next up previous contents
Next: Parabon Up: Peer to Peer Computing Previous: Javelin   Contents

SETI@home

SETI@home [52], a project that searches for extraterrestrial intelligence using networked personal computers, was launched early 1998. The Internet clients analyze data that is collected by the Arecibo radio telescope looking for possible indication of extraterrestrial intelligence. The collected data is divided into work units of 0.25 Mbyte. A work unit is big enough to keep a computer busy for a while and small enough to transmit in a few minutes even for a 28.8Kbps modem. The essential part of this project is the client program, which is a screen saver for Windows or Macintosh users. Hence the client program will run only when computer is not being used. A Sign-up client gets a work unit from SETI@home data distribution server. After finishing processing its data, the client sends results back to the server and gets a new work unit. The SETI@home data server connects only when transferring data. To protect the job from computer failure, the client program writes a ``check point'' file to disk. Hence the program can pick up where it left off. The SETI@home data distribution server keeps track of the work units with a large database. When the work units are returned, they are put back into the database and the server looks for a new work unit and sends it out. Each work unit is sent out multiple times in order to make sure that the data is processed correctly. SETI@home is faster then ASCI White, which is currently the fastest supercomputer, having peak performance of $ 12.3 \times 1024$ floating-point operations per seconds (TFLOPS). 3.1 trillion floating-point operations are required by a work unit to compute an FFT. SETI@home clients (between them) process about 700,000 work units in a typical day. This works out to over 20 TFLOPS. SETI@home also costs less then 1% compared to ASCI White. In the first week after the launch, over 200,000 people downloaded and ran the client. This number has grown to 2,400,000 as of October 2000. People in 226 countries around the world run SETI@home. You can find more information about this project and client sign up are available from http://setiathome.ssl.berkeley.edu. The SETI@home approach is not suitable for all problems. It must be possible to factor the problem into a large number of pieces that can be handled in parallel, with few or no interdependencies between the pieces.
next up previous contents
Next: Parabon Up: Peer to Peer Computing Previous: Javelin   Contents
Bryan Carpenter 2004-06-09