OWS: Report on the recently completed batch
Verfasst: 23.02.2015 11:46
Hi everyone,
Thank you so much for your support to yoyo@home and to Odd Weird Search! We just completed the current batch on numbers lower than 10^28 with excess (or abundance) at most 10^14. In this batch, we released about 230000 workunits, and we have checked 4323850414816 = 4.3*10^12 abundant numbers for their "weirdness". This will take about 150 years on a single core, so we could not possibly accomplish it without you!
Although we didn't find any odd weird number in our search, which goes against our expectation, but to quote Rajesh Koothrappali, "Negative result is still a result". Even if we didn't look at every number below 10^28, we have looked at the most probable place that an odd weird number will be. So future explorers can save their time and jump directly to a range higher than 10^28.
For the future development of OWS, there are a few problems that we need to solve before any new batch.
One problem is workunit size. Most of the OWS workunits are of moderate size (about 2 hours), but some very nasty ones can takes days to complete, and they incur a lot of problems among you crunchers. We have identify the problem, but the solution is still to be implemented and tested, and it may involve a rewrite of the main loop.
The other problem is the structural overhead. We notice that the current batch is larger than the previous one, but the number of abundant numbers checked is much lower. This is because we have to explore the multiplicative structure of natural numbers to find our candidates, and this exploration has its computational cost, and the cost is significantly larger in the current one than the previous one. We now want to find a way to lower this "structural" cost, since it goes higher for larger numbers. If we want to do, for example, 10^33 with 10^11 abundance, we will have to deal with this problem.
We are still in the discussion, so your opinion is welcomed. If anyone is interested in the development, or just want to see the code and raw data, please PM me for a discussion.
Cheers,
fwjmath.
Thank you so much for your support to yoyo@home and to Odd Weird Search! We just completed the current batch on numbers lower than 10^28 with excess (or abundance) at most 10^14. In this batch, we released about 230000 workunits, and we have checked 4323850414816 = 4.3*10^12 abundant numbers for their "weirdness". This will take about 150 years on a single core, so we could not possibly accomplish it without you!
Although we didn't find any odd weird number in our search, which goes against our expectation, but to quote Rajesh Koothrappali, "Negative result is still a result". Even if we didn't look at every number below 10^28, we have looked at the most probable place that an odd weird number will be. So future explorers can save their time and jump directly to a range higher than 10^28.
For the future development of OWS, there are a few problems that we need to solve before any new batch.
One problem is workunit size. Most of the OWS workunits are of moderate size (about 2 hours), but some very nasty ones can takes days to complete, and they incur a lot of problems among you crunchers. We have identify the problem, but the solution is still to be implemented and tested, and it may involve a rewrite of the main loop.
The other problem is the structural overhead. We notice that the current batch is larger than the previous one, but the number of abundant numbers checked is much lower. This is because we have to explore the multiplicative structure of natural numbers to find our candidates, and this exploration has its computational cost, and the cost is significantly larger in the current one than the previous one. We now want to find a way to lower this "structural" cost, since it goes higher for larger numbers. If we want to do, for example, 10^33 with 10^11 abundance, we will have to deal with this problem.
We are still in the discussion, so your opinion is welcomed. If anyone is interested in the development, or just want to see the code and raw data, please PM me for a discussion.
Cheers,
fwjmath.