4-5M Reservations - Page 2

jrk · 2008-06-03, 01:52

Oh I don't doubt that double-checking is important (I do it occasionally to test hardware). I just wonder why it can't be a few weeks behind first-time testing, you know, so if someone's interested in making the record list he can make it without getting demoted as a double-check.

But maybe that doesn't matter. I dunno. :)

paulunderwood · 2008-06-03, 02:16

That might be a good idea. I do not know PG's error rates for 321.

On a similar note, wouldn't it be fun to start where we expect the next prime to be and to work outward from that until one is found and then repeat the operation again?

jrk · 2008-06-03, 03:09

Yep. It'd also be fun to pick your own assignments, too, if that were possible. So people could entertain their own ideas (though, just as a game, you can't really say much about the distribution of these numbers anyway

).

Rytis · 2008-06-05, 18:01

There is a new feature coming into BOINC, which will assign a reliability score to computers (they will first crunch doublechecked workunits, and when it's known that the host produces valid data, it will be issued non-doublecheck work). The plan is to have only about 10-20% of work doublechecked and still ensure valid results. And we can always go back to full doublecheck if needed or revisit the single-check numbers later.

Over the last 7 days we had 6.6537% fail rate on Linux (1.1305% aborted by user, 0.2261% returned wrong residue, total of 3096 tests completed) and 11.6169% fail rate on Windows (5.1209% user aborts, 1.0748% wrong residue, total of 12654 tests completed).

Kosmaj · 2008-06-08, 03:29

People who like the prime testing style of 321 are welcome to joing our Riesel Prime Search project searching for primes of form k*2^n-1. We have manual reservations and use LLR only. Everybody is free to test the numbers he/she likes.

Well sieved values in the megabit prime region (n>1M) are available right now for k=5, 15, 17, and 105. Presieved candidates for many k<300 and n<1M organized in so-called "drives" are available as well. Feel free to drop by and have a look.

Thanks,
Kosmaj

2008-06-03, 02:16	#13
paulunderwood Sep 2002 Database er0rr 7×23×29 Posts	That might be a good idea. I do not know PG's error rates for 321. On a similar note, wouldn't it be fun to start where we expect the next prime to be and to work outward from that until one is found and then repeat the operation again? Last fiddled with by paulunderwood on 2008-06-03 at 02:17

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Reservations	ET_	Operazione Doppi Mersennes	561	2023-05-27 13:15
Reservations	kar_bon	Riesel Prime Data Collecting (k*2^n-1)	136	2023-02-14 06:58
Where are my reservations	fivemack	PrimeNet	3	2016-02-08 17:58
Reservations?	R.D. Silverman	NFS@Home	15	2015-11-29 23:18
Reservations	R.D. Silverman	Cunningham Tables	15	2011-03-04 21:01

2008-06-03, 01:52	#12
jrk May 2008 3×5×73 Posts	Oh I don't doubt that double-checking is important (I do it occasionally to test hardware). I just wonder why it can't be a few weeks behind first-time testing, you know, so if someone's interested in making the record list he can make it without getting demoted as a double-check. But maybe that doesn't matter. I dunno. :)

2008-06-03, 03:09	#14
jrk May 2008 2107₈ Posts	Yep. It'd also be fun to pick your own assignments, too, if that were possible. So people could entertain their own ideas (though, just as a game, you can't really say much about the distribution of these numbers anyway ).

2008-06-05, 18:01	#15
Rytis Nov 2006 2³×11 Posts	There is a new feature coming into BOINC, which will assign a reliability score to computers (they will first crunch doublechecked workunits, and when it's known that the host produces valid data, it will be issued non-doublecheck work). The plan is to have only about 10-20% of work doublechecked and still ensure valid results. And we can always go back to full doublecheck if needed or revisit the single-check numbers later. Over the last 7 days we had 6.6537% fail rate on Linux (1.1305% aborted by user, 0.2261% returned wrong residue, total of 3096 tests completed) and 11.6169% fail rate on Windows (5.1209% user aborts, 1.0748% wrong residue, total of 12654 tests completed).

2008-06-08, 03:29	#16
Kosmaj Nov 2003 2×1,811 Posts	People who like the prime testing style of 321 are welcome to joing our Riesel Prime Search project searching for primes of form k*2^n-1. We have manual reservations and use LLR only. Everybody is free to test the numbers he/she likes. Well sieved values in the megabit prime region (n>1M) are available right now for k=5, 15, 17, and 105. Presieved candidates for many k<300 and n<1M organized in so-called "drives" are available as well. Feel free to drop by and have a look. Thanks, Kosmaj