Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Sieving :
Why have the sieving stats been crippled?
| Author |
Message |
|---|
GDBSend message
Joined: 15 Nov 11
Posts: 298
ID: 119185
Credit: 4,064,951,190
RAC: 1,955,120
                      
|
|
The manual sieving stats used to show a lot of useful info for each sieving range. Now, the stats have been stripped down to only showing factors removed. How come? | |
|
JimB Honorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,270,144
RAC: 179,433
              
|
The manual sieving stats used to show a lot of useful info for each sieving range. Now, the stats have been stripped down to only showing factors removed. How come?
Several reasons. The old stats used to show time saved using both the OCL2 and OCL3 transforms based on their respective b-ranges. OCL2 is no longer used and now the various transforms have overlapping ranges and the choice of which is used depends on the individual computer. So the old data had become entirely fictitious and had no relation to reality. Without being able to tell which transform would be used on any particular candidate, it's not possible to generate an accurate time saved value. And even on the old stats, the numbers were not real because they made no distinction between candidates already tested by genefer (so no time was saved) and those sieved out before genefer got that high.
Because now we're sieving GFN15 and GFN16 to b=2G instead of b=100M the factor files are 20 times as large. I've chosen to show stats for GFN15 and GFN16 to b=400M because that's the highest b value that genefer can test. Those time saved fields were not some kind of average, they were computed individually for each candidate removed and then totaled. Trying to do that for four times the number of candidates was slower and was taking up too much room on my system (each candidate eliminated had its time to compute calculated once and then that time was stored). Plus even though I'm only providing numbers up to b=400M all the other candidates up to b=2G were still in the files slowing everything down.
For the same kinds of reasons I removed the graph of candidates removed from the sieve and kept only the factors found graph. Keeping track of that many individual candidates removed and graphing four times as many vertically meant that it took about an hour to generate that graph for any single n. Before this I could generate graphs for GFN15, 16, 17, 21, 22 simultaneously and have them all within five minutes. Given that I may process pending sieving 5-6 times a day I can't afford to have the graph generation take an hour each time. And remember that this is all taking place on my home computers, where there's quite a bit of infrastructure devoted to it (currently 22 databases relating to PrimeGrid and sieving).
Bottom line: I didn't want to lose those extra stats any more than you, but they just weren't practical any more. My time and CPU are better spent elsewhere. | |
|
GDBSend message
Joined: 15 Nov 11
Posts: 298
ID: 119185
Credit: 4,064,951,190
RAC: 1,955,120
|
|
Since the old sieving stats method can't be reasonably performed, can a new method be found to show how many "useful" factors were removed by sieving. A "useful" factor might be a number that would've been GFN or LLR tested within the next 10 years, but was found and removed by sieving first.
If this can be done easily, it would provide some measure of how useful manual sieving is. | |
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,270,144
RAC: 179,433
|
|
I'll have to think about it. There are practical problems I'd have to deal with and right now I'm putting in the majority of my time on other things. So this will go on the back burner for now.
| |
|
GDBSend message
Joined: 15 Nov 11
Posts: 298
ID: 119185
Credit: 4,064,951,190
RAC: 1,955,120
|
|
Is it possible to get a count of probable factors not removed from sieve for n=22 & b=80,000 thru 200,000? This range covers my estimate to 10 years of GFN-22 searching. If we can get the factor count again in a month or so, we can estimate how many 10 year factors are removed per P sieved. From this we can determine how effective our sieving is. Thanks! | |
|
Message boards :
Sieving :
Why have the sieving stats been crippled? |
