The project organizers have also made it possible for participants to pool their results in Teams, which are in turn organized under categories like Large Company, Small Company, Primary Education, Club Teams, and several other categories. The Mac Observer is proud to have the #15 Club Team with 712 members who have contributed 441.15 CPU years. We would like you to join our Team too, and help show the world how fast the Mac is. The Mac platform has been producing Work Units far faster than has the Pentium/Windows platform from the beginning, and Team Mac Observer needs your help to keep on doing that. You can find more information on the Team and [email protected] at our Team Mac Observer Web page.
The new version of the [email protected] client allows SETI researchers to search for low orbit satellites, similar to those in orbit around planet Earth. It also adds a number of other new calculations, optimized performance, and other improvements. According to the SETI team:
Changes in version 3.0 of [email protected]:
- Optimization of the FFT computation. We have adopted the highly optimized Ooura FFT routines developed at the University of Tokyo. Our tests show a 60% client speed up over version 2.x with a a 4% sacrifice in memory consumption.
- With the CPU time saved by optimizing the FFT code, we were able to add a major new scientific component. The [email protected] client now performs two types of pulse detection. This means that you will occasionally see two new signal types in your result file: "pulse" and "triplet". These algorithms are fully described here.
- We were also able to increase the range of doppler drift rates at which the client analyzes the each work unit. We increased the range from +- 10Hz/s to +- 50Hz/s. At this range, we could detect an extraterrestrial version of our own low earth orbiting satellites. This is a search space never before explored. In order to not extend execution time too much, we only process the data at .3Hz and coarser spectral resolution for drift rates beyond +- 10Hz/second.
- Improvement in state saving. The pulse detection algorithm is very compute intensive at coarse spectral (fine time) resolutions. We extended the state saving algorithm so that on restart the client will return to where it was in gaussian/pulse/triplet finding rather than starting over on the current doppler drift rate and spectral resolution.
- We made our gaussian fit (weighted chi-square) threshold more strict by lowering it from 10.0 to 8.8. This was done for two reasons. First, when we looked a number of gaussian results, we saw that those with fits between 8.8 and 10.0 did not exhibit a convincing gaussian shape. Instead, they appeared to more spike like. Second, with the extension of the doppler drift range, our gaussian return rate would go up by a factor of five. Given that the majority of these would be false positive, we tightened up the threshold.
- New graphics display the pulse detection algorithms.
- The panel that displays in-progress gaussian/pulse/triplet searching cycles among the best-of-workunit of each of these signal types when the client is not performing one of these searches.
Note on execution time and individual/team statistics
The combined effect of FFT optimization, pulse detection, and the extended doppler drift range is that a typical workunit will take about 40% longer to complete with version 3.0 on any given platform. On balance, this gives the project the best science for the CPU cycles used. We trimmed processing where we could to lessen the impact on execution time. This of course affects the stats. It will take longer to add to your results received statistic. While it might be nice to somehow make version 3.0 workunits "worth more" stats wise, we have not added that complexity.