mersenneforum.org Wildly differing times for step 2 with ecm-6.0.1
 Register FAQ Search Today's Posts Mark Forums Read

 2005-09-07, 21:33 #1 Jushi     Sep 2005 UGent 22·3·5 Posts Wildly differing times for step 2 with ecm-6.0.1 I recently started exploring GMP-ECM and noted that for a given number and given parameters, the Step 1 time is almost constant, but Step 2 time varies from run to run up to a factor 5 or so. This is on one and the same machine for 5 curves: Code: Step 1 took 1188420ms Step 1 took 1188460ms Step 1 took 1189800ms Step 1 took 1188310ms Step 1 took 1189760ms Code: Step 2 took 671880ms Step 2 took 718610ms Step 2 took 2851240ms Step 2 took 475260ms Step 2 took 558580ms Is this normal or am I doing something wrong?
 2005-09-07, 21:38 #2 akruppa     "Nancy" Aug 2002 Alexandria 2,467 Posts Stage 2 can use quite a lot of RAM. Is the computer running out of memory and starting to swap to disk a lot? In that case, the performance you'd get would depend i.e. on how many other applications are running. What system are you on, what number are you factoring and with what parameters? Alex
 2005-09-10, 08:11 #3 Jushi     Sep 2005 UGent 22×3×5 Posts Looking at a top', I think I can explain the problem: a load average of 1.3 (should be 1.0). However, apart from ecm' I see nothing else running... so that will require some investigation. By the way: the machine is a PentiumIV 3.0GHz with 2GB ram and zero swap. The number is a C288, and I'm running ecm-6.0.1 with B1=11e6. Other question: Is there any difference between running `ecm' ten times, as opposed to running once with a -c 10 option?
2005-09-10, 09:23   #4
geoff

Mar 2003
New Zealand

13·89 Posts

Quote:
 Originally Posted by Jushi By the way: the machine is a PentiumIV 3.0GHz with 2GB ram and zero swap. The number is a C288, and I'm running ecm-6.0.1 with B1=11e6.
Those stage one times (and the fastest stage two time) are almost double what a number that size should require on that machine.

2005-09-11, 21:25   #5
Jushi

Sep 2005
UGent

22×3×5 Posts

The problem with differing times seems to have resolved itself by rebooting (I still don't know what caused it).

Quote:
 Originally Posted by geoff Those stage one times (and the fastest stage two time) are almost double what a number that size should require on that machine.
Are you sure? What else could influence the running time...?

2005-09-11, 23:19   #6
Mystwalker

Jul 2004
Potsdam, Germany

3·277 Posts

Quote:
 Originally Posted by Jushi Are you sure? What else could influence the running time...?
Maybe another application that takes "full" processing power, in case you have Hyperthreading enabled?

2005-09-12, 00:33   #7
geoff

Mar 2003
New Zealand

13×89 Posts

Quote:
 Originally Posted by Mystwalker Maybe another application that takes "full" processing power, in case you have Hyperthreading enabled?
Another possibility is that you are using a GMP library optimised for the wrong processor. If that is the case then recompiling with P4-specific optimisations could give a very large improvement. (For example, the precompiled Debian library is optimised for a 486, recompiling for a P4 makes ecm run more than 70% faster).

 2005-09-12, 01:30 #8 geoff     Mar 2003 New Zealand 13×89 Posts As a reference point, here is a curve for a general 284 digit number on a 2.66GHz P4: GMP-ECM 6.0 [powered by GMP 4.1.4] [ECM] Input number is 1659 ... 8037 (284 digits) Using B1=11000000, B2=25577181640, polynomial Dickson(12), sigma=3150013712 Step 1 took 645928ms Step 2 took 208735ms

 Similar Threads Thread Thread Starter Forum Replies Last Post lycorn Lone Mersenne Hunters 37 2011-10-09 16:09 CRGreathouse Factoring 9 2010-10-30 16:38 Andi47 GMP-ECM 9 2007-04-12 11:40 alpertron Software 4 2006-01-11 17:27 Ken_g6 Puzzles 1 2005-01-16 15:03

All times are UTC. The time now is 16:29.

Sat May 21 16:29:18 UTC 2022 up 37 days, 14:30, 0 users, load averages: 1.87, 1.48, 1.36