Posts by bdodson*
41)
Message boards :
NFS Discussion :
Very long post processing
(Message 249)
Posted 26 Nov 2009 by bdodson* Post: Greg, who is doing most of the post processing? What type of machines/cluster are being used? Just curious. Greg's always done most of the post processing. Only the intensive sparse matrix calculation has been farmed out on recent numbers. One of my other friends reports having done one of the recent November matrices, and waiting for R269 (a number of larger "difficulty", with a matrix that will take longer). I'd be interested to hear more, as well, but am not sure how soon Greg will get back online with the local holiday(s). -Bruce |
42)
Message boards :
NFS Discussion :
3,538+ factors
(Message 218)
Posted 6 Nov 2009 by bdodson* Post: The 177-digit number we factored was the last composite factor, the other known factors all being primes; so now this last piece has been reduced to primes, and we're done with this number. The one we're going on to is the next one on the "Status of Numbers" list, ordered by completion date, 2,853+. Hope this clarifies matters. -bdodson |
43)
Message boards :
Questions/Problems/Bugs :
Project recently started using too much memory?
(Message 216)
Posted 6 Nov 2009 by bdodson* Post: ...select "If no work for selected applications is available, accept work from other applications" so if lasievef does run out of work, then your computer won't be idle. Done, thanks. Just to be clear, I'm still banning the 16e siever ("lasievef") from linux machines with 1Gb/core or less memory, as well as on the Windows machines. Task Manager reports 356,172K for 15e ("lasievee"), but there's no reason to risk having boinc being in the way. -bdodson |
44)
Message boards :
Questions/Problems/Bugs :
Project recently started using too much memory?
(Message 214)
Posted 6 Nov 2009 by bdodson* Post: ... BOINC allows you to designate profiles called home, school, and work. You can, say, call the computers without sufficient memory "school" computers by viewing the computers on your account, selecting details for the low memory computers, then changing location near the bottom. Then, in your NFS@Home preferences you can add separate preferences for school and disable lasievef ... On our Xeon cluster(s) the memory use for lasievef on the new target R269 is nowhere near 1Gb: 19 708m 330m 724 R 100.1 0.7 4:51.67 lasievef_1.07_x 19 627m 322m 720 R 100.1 0.7 4:45.84 lasievee_1.07_x which shows lowest priority 19, then virtual memory, then RAM used. I'm switching these from "school" to "work", with the latter profile _only_ accepting lasievef tasks. If sufficiently many other users/computers are set like this, Greg may have to monitor the available tasks (in the server settings) to make sure that we don't run out of lasievef tasks. This hasn't been a problem with few people selecting the "large" or "huge" setting. I'm especially interested in the factors of R269, a number with larger "difficulty" (on the "Status of Numbers" page). These numbers get more extensive ECM pretests, and keeping numbers requiring the 16e siever feasible gives us a larger pool of interesting candidates. -bdodson |
45)
Message boards :
NFS Discussion :
1M credits/day
(Message 178)
Posted 24 Oct 2009 by bdodson* Post: I mean overall. We all together are at 727k credits per day. I started from Stats & Leaders, clicked on BOINC Stats, clicked to get the full list (we're not in the top15; is that a plausible objective? We'd certainly raise our profile ...). Then I clicked on the column for "last day". The NFS@Home line is still saying 1.13M, with a last XML update of 5 hours ago (8 GMT). Where should I be looking for the 727k? -bdodson PS -- The top15 in question is by "most active", which appears to mean the "last day" column. If I'm reading correctly, we're in 16th, just a bit behind cosmology and "spinhenge". |
46)
Message boards :
NFS Discussion :
1M credits/day
(Message 167)
Posted 24 Oct 2009 by bdodson* Post: Can we manage 1M credits per day? You mean like this? NFS@Home 702 +15 1,962 +40 129 +2 51 0 18,315,333 +1,130,532 02:33:02 old (looks like 1.13M to me!) -bd |
47)
Message boards :
NFS Discussion :
Computer Lockup's
(Message 163)
Posted 24 Oct 2009 by bdodson* Post: ... Our problem is that (1) there are very few Cunningham numbers for which the low memory siever is the correct choice; and (2) projects for numbers that size would finish very quickly, just 2-3 days, which is already difficult to manage for the medium-sized siever (as illustrated by the 59-digit ECM factor in the most recent reservations). I could be wrong, but I don't believe that any of our numbers have used the small-memory siever. I've recently been sieving a bunch of numbers with difficulty 240.0-249.99 on our x86-64 clusters (distributed under condor) --- they also use the medium-sized siever. The NFS@Home numbers all have difficulty 250.0-259.99, still in the range where the medium-sized siever is best. I did one number with the large siever which had difficulty 269, but I'm not sure where the crossover is (I'm not sure that we know). I've only done one number with the low-memory siever, which had difficulty in the 220's. One of our friends is expert at finding interest in those small-sized numbers (as well as providing the assembly code settings used for the large memory siever, which is new). I'll check to see whether there's one or two that we could convince Greg would be worth the over-head. Sorry to hear that the new tasks crashed your machine. We had some discussions with the leader of yoyo@home, who was assuring us that BOINC would be able to track larger memory jobs away from machines on which they'd cause a problem. Sounds like we weren't careful enough. Regards, bdodson |
48)
Message boards :
Questions/Problems/Bugs :
More info on "Status of numbers" page
(Message 154)
Posted 22 Oct 2009 by bdodson* Post: Good idea. We are currently getting about one result per week now, and the date will help monitor that rate. There seems to still be space on the status page, perhaps we could have the size of the prime factors? One of the things SNFS factorization is good for is that it checks the performance of ECM pretesting (it's hardly ever ECM "factoring", in this range). Too many factorizations where the smallest prime factor was out of ECM range, and it's hard to keep pushing up the effort. While an "ECM miss" (a 53-digit prime factor, or a 54-digit prime factor, after sufficient testing to remove 55-digit primes to 80%) is less expensive than a miss in GNFS. And yes, Greg's fast. -bdodson Cunningham "Champion" SNFS factorizations with Greg:
|
49)
Message boards :
NFS Discussion :
p59 factor of 6, 334+ by ecm
(Message 141)
Posted 20 Oct 2009 by bdodson* Post: The early NFS@Home numbers were all tested extensively by ECM (the Elliptic Curve Method) for small and medium sized prime factors, past 55-digits. The three most recent numbers were less tested, and of these 6, 334+ was hardly tested at all before selection (due to rapid NFS@Home progress!). We had expected 50-digit prime factors to have been removed, and planned on removing prime factors up to 55-digits, with a chance of finding larger factors, before starting sieving. So I'm happy to report that the 59-digit prime p59 = 37597376323754357344197406664995834047249702145969970498293 divides the 239-digit composite factor of 6,334+ we had intended to factor. Unfortunately, there remains a composite cofactor of 181-digits; and sieving by snfs (with difficulty 259) is still easier than using gnfs (with gnfs on a 181-digit number perhaps comparable in difficulty to snfs on a number with difficulty of 270-digits). I'm not done testing yet, but there's most likely not another factor in ECM range (up to 70-digit primes). This factor was found by GMP-ECM, under the ECMNET project, Dodson/ECMNET; which is also the software used by the boinc project yoyo (for which Beyond of Team Ars Technica found a recent large prime factor). The machine used was one of 300+ 32-bit linux/xeons, along with c. 500 pcs (1050 pcs used during 8pm to 8am) distributed under condor. The first step limit B1 was 260,000,000 (with default B2), optimal for finding 60-digit primes. This p59 is the 4th largest prime found this year by ECMNET, just below the 59-digit prime in 3rd place found by yoyo@home earlier this year. It's nowhere near the Top 10 on Brent's list, for which the three smallest primes all have 62-digits. At the risk of trying the patience of NFS@home readers, I can report that four of those Top 10 are Dodson/ECMNET factors, including the two largest at 67-digits and 66-digits (just a bit larger than the 66-digit prime found by the ECMNET founder, Paul Zimmermann). -bdodson |
50)
Message boards :
NFS Discussion :
Reservations?
(Message 140)
Posted 19 Oct 2009 by bdodson* Post: The three most recent additions to the list under "Status of Numbers" are the Cunningham numbers of "snfs difficulty" 259 (neglecting two at 258.8 that would have required a quartic polynomial). The two you mention are at 252.5; so more overhead switching numbers. One of the three added is a 2LM (1726L), so 2LM's aren't being neglected here (two of the 16 on the "Numbers" list). One of the others just added is the 3rd-from-last to be reserved from the previous list of "Wanted" numbers (viz, 7, 307+; with the other two above the current NFS@Home range). From the new "Wanted" lists Greg's post refers to (mailed as hardcopy with the factors on Sam's "Page 112"; we're still waiting for the online version), all of the "Most Wanted" are already reserved (three currently here in "Post Processing"). Not counting the third of our new "Numbers" (6, 334+), there are just ten new "More Wanted" not already reserved, including the two you've mentioned. We should be able to clear several of these. -bdodson (PS - The term "snfs difficulty" refers to the estimate for the runtime of our factorizations, as measured by the asymptotic formula.) |