mersenneforum.org gmp-ecm crash
 Register FAQ Search Today's Posts Mark Forums Read

2011-01-18, 16:19   #1
yoyo

Oct 2006
Berlin, Germany

62610 Posts
gmp-ecm crash

Hello,

my yoyo@home users have probems with this C35842 and gmp-ecm.
gmp-ecm runs with this command line:
Quote:
 ecm.exe -v -nn -timestamp -chkpnt checkpnt -inp in -k 8 -maxmem 1800 250000
Here is the output of one of the crashing tasks of a 64 bit Windows 7 host.

1) It seems that gmp-ecm allocates more memory than specified with -maxmem. We saw up to 2200 before it crashed.
Why? Is the combination of parameters wrong?

2) On Windows 7/xp 64/32 bit and Linux 32 bit gmp-ecm crashes. Only Linux 64 is able to run a curve on this number. Even if we increase -maxmem it crashes on these systems. Any Idea here? Can somebody test or confirm it?

yoyo

 2011-01-18, 21:23 #2 ATH Einyen     Dec 2003 Denmark 61738 Posts I think you are supposed to use -k or -maxmem but not both at once. See "README" section "4. Memory usage.". Your -k 8 probably overrules -maxmem 1800 so it uses too much memory. But I'm far from any expert on the subject, so I might be wrong. Last fiddled with by ATH on 2011-01-18 at 21:23
 2011-01-19, 18:03 #3 henryzz Just call me Henry     "David" Sep 2007 Cambridge (GMT/BST) 10111001100012 Posts I ran on win64 and stage 1 worked. It found the factor 627986403 in stage 1.
2011-01-19, 18:34   #4
yoyo

Oct 2006
Berlin, Germany

2·313 Posts

Oh, seems that I provided somehow the wrong input number. Some copy&paste fault. The number has 65050 digits. I updated the file.

We were running it without -k and -maxmem options and it crashed also. Just before the crash it was using 3.3GB ram. Last line in output was:
Quote:
 .................5576029195936544660 * x^2879) Building G from its roots took 308461ms
yoyo

2011-01-19, 23:03   #5
ATH
Einyen

Dec 2003
Denmark

1100011110112 Posts

If you removed both -k and -maxmem there is no restraint on memory use, so it will crash attempting to use too much.

Try removing the -k 8, I just tried it and didn't crash for me, max memory usage was around 1860 Kb:

Quote:
 ecm.exe -v -nn -timestamp -chkpnt checkpnt -inp in -maxmem 1800 250000

2011-01-20, 08:26   #6
yoyo

Oct 2006
Berlin, Germany

2·313 Posts

Quote:
 Originally Posted by ATH If you removed both -k and -maxmem there is no restraint on memory use, so it will crash attempting to use too much. Try removing the -k 8, I just tried it and didn't crash for me, max memory usage was around 1860 Kb:
We tried even this and it crashed and even without the -maxmem option it crashed after consuming 3.7GB ram.

Summary:
-k 8 -maxmem 1800 -> crash
-maxmem 1800 -> crash
<without both> -> crash

It crashed always in step 2 after:
Code:
Building G from its roots took 130558ms
Computing G * H took 30576ms
Tests were done with gmp-ecm 6.2.3 and 6.3
Runtime on a Win64 system was ~1h.
yoyo

 2011-01-21, 00:00 #7 ATH Einyen     Dec 2003 Denmark 61738 Posts It worked for me with this gmp-ecm 6.3-rc4 64bit. Compiled with Msys+Mingw64 on WinXP: ecm63rc4-core2-64.zip or maybe try the gmp-ecm 6.3 I just compiled here: http://www.mersenneforum.org/showpos...&postcount=213
 2011-01-21, 05:11 #8 yoyo     Oct 2006 Berlin, Germany 11628 Posts We will try your version. We used before these versions http://gilchrist.ca/jeff/factoring/ yoyo
 2011-01-21, 21:08 #9 yoyo     Oct 2006 Berlin, Germany 2·313 Posts Your Win64 versions are running fine. But they seem to be much slower than these from Jeff. Unfortunately Jeff's version are crashing for this big number. Do you have an explanation for it? All testresults are here http://www.rechenkraft.net/phpBB/vie...11438&start=24 yoyo
 2011-01-22, 18:35 #10 ATH Einyen     Dec 2003 Denmark 32×5×71 Posts You are right. I did some tests: GMPECMtests.html Up to 350 digits Jeff's binary and my binary with asm-redc is equally fast. From 350 to 550 digits Jeff's binary is much faster than mine, but above 550 digits Jeff's binary crashes for me too in stage2. I don't know what is the reason for this big speed difference. It must be either that MPIR is faster than GMP for bigger numbers or maybe because he compiled it with Visual Studio while I compiled with Mingw64 gcc. On a side note the binary with asm-redc is faster than the one without up to ~600 digits. Then from 600 up to 2000 digits they are about equal speed. Note: This is for 64bit systems, last time I tested a 32bit binary the one without asm-redc was faster from ~200 digits and up.
 2011-01-23, 01:32 #11 ATH Einyen     Dec 2003 Denmark 32·5·71 Posts It seems to be MPIR making it faster. I compiled my own version with MPIR 2.1.1 and its even alittle faster than Jeff's binary, get it here: http://www.mersenneforum.org/showpos...&postcount=215 With or without --enable-asm-redc seems equally fast at least up to 2000 digits. Last fiddled with by ATH on 2011-01-23 at 01:32

 Similar Threads Thread Thread Starter Forum Replies Last Post lavalamp GMP-ECM 55 2011-04-03 01:58 storm5510 Software 8 2009-08-31 02:07 yoyo GMP-ECM 2 2009-01-26 19:22 remaker Software 4 2007-05-03 16:39 coastertux Software 2 2004-03-16 20:49

All times are UTC. The time now is 21:59.

Mon Nov 29 21:59:39 UTC 2021 up 129 days, 16:28, 1 user, load averages: 1.15, 1.47, 1.51