Too much time spent on ECM

jux · 2016-01-03, 04:29

Probably configuring something wrong again but YAFU decided to spend 10 hours on ECM for a C134
Is this normal?

Dubslow · 2016-01-03, 04:40

If it's single threaded, probably.

On my i7-2600K, it's roughly 1-2 hours of ECM with 8 threads, before maybe ~12 hours of NFS with 8 threads. (These time estimates are order of magnitude.)

jux · 2016-01-03, 04:52

I ran it with 8 threads though
Old version of YAFU (1.24) never spent this long on ECM

VBCurtis · 2016-01-03, 04:56

How many curves at what size did it run for the last phase of ECM?

I haven't used YAFU in a while, but there is a setting in the .ini to control the ECM depth. I think it's set to 0.33 by default; you can change this to 0.31 to significantly reduce ECM effort for each composite.

0.33 * 134 = 44 and change, so it likely ran most of the B1 = 11M level, 3000 curves or more.
0.31 * 134 = 41.5, which is something like 500 curves at 11M (after the same number of curves at 3M as the 0.33 setting).

Dubslow · 2016-01-03, 06:17

For GNFS, the default is 4/13s the GNFS difficulty.

Also, are you sure that there were actually 8 threads of ECM running? I seem to recall some difficulties about that recently.

jux · 2016-01-03, 07:53

Code:

Finished 541 curves using Lenstra ECM method on C134 input, B1=11M, B2=gmp-ecm default

Though during ECM CPU usage was only around 20%
I don't think it was really running with 8 threads

VBCurtis · 2016-01-03, 08:01

the complete set of curves at 3M plus 541 curves at 11M won't take 2 hours with 8 threads, so it definitely wasn't running multi-threaded for the ECM phase.

I wouldn't change the ratio- 541 curves @ 11M for that number sounds pretty good to me; I either last ran YAFU when the default ratio was higher, or I am simply mistaken.

jux · 2016-01-03, 08:45

I have always considered ECM to be a quick check to remove some small factors - I don't want it running for more than an hour
Any ideas on how to get it fully multithreaded?

Dubslow · 2016-01-03, 08:55

Quote:

Originally Posted by jux

I have always considered ECM to be a quick check to remove some small factors - I don't want it running for more than an hour

Yafu and factmsieve.py have been relatively carefully tuned over many years (decades, really) of experience by very smart people well informed on the topic. They are very nearly optimal in the amount of ECM they run (at least when NFS and ECM use the same number of threads). Unless you have some experience under your belt and know exactly what you're doing, you're far better off just letting them run howevermuch ECM they deem necessary.

Quote:

Originally Posted by jux

Any ideas on how to get it fully multithreaded?

This is the crux of course. As I've mentioned elsewhere, yafu's builtin ECM is incapable of being multithreaded -- however if you download GMP-ECM and point yafu to that executable (via yafu.ini), then ECM will be multithreaded.

LaurV · 2016-01-03, 15:09

Quote:

Originally Posted by jux

I have always considered ECM to be a quick check to remove some small factors - I don't want it running for more than an hour

you can use "pretest_ratio=" in the ini file to set the amount of the ecm to be done

jux · 2016-01-03, 22:00

Quote:

Originally Posted by Dubslow

This is the crux of course. As I've mentioned elsewhere, yafu's builtin ECM is incapable of being multithreaded -- however if you download GMP-ECM and point yafu to that executable (via yafu.ini), then ECM will be multithreaded.

Didn't realize that
I've added it now

2016-01-03, 04:29	#1
jux Aug 2015 2·3³ Posts	Too much time spent on ECM Probably configuring something wrong again but YAFU decided to spend 10 hours on ECM for a C134 Is this normal?

2016-01-03, 04:56	#4
VBCurtis "Curtis" Feb 2005 Riverside, CA 1507₁₆ Posts	How many curves at what size did it run for the last phase of ECM? I haven't used YAFU in a while, but there is a setting in the .ini to control the ECM depth. I think it's set to 0.33 by default; you can change this to 0.31 to significantly reduce ECM effort for each composite. 0.33 * 134 = 44 and change, so it likely ran most of the B1 = 11M level, 3000 curves or more. 0.31 * 134 = 41.5, which is something like 500 curves at 11M (after the same number of curves at 3M as the 0.33 setting). Last fiddled with by VBCurtis on 2016-01-03 at 04:57 Reason: typo

2016-01-03, 06:17	#5
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3·29·83 Posts	For GNFS, the default is 4/13s the GNFS difficulty. Also, are you sure that there were actually 8 threads of ECM running? I seem to recall some difficulties about that recently. Last fiddled with by Dubslow on 2016-01-03 at 06:18

2016-01-03, 07:53	#6
jux Aug 2015 2·3³ Posts	Code: Finished 541 curves using Lenstra ECM method on C134 input, B1=11M, B2=gmp-ecm default Though during ECM CPU usage was only around 20% I don't think it was really running with 8 threads

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
P95 est time to go seems off	bcp19	Software	1	2012-08-03 22:39
How much time	Unregistered	Information & Answers	4	2008-12-20 21:00
New .dat time?	benjackson	Prime Sierpinski Project	16	2008-07-29 07:26
Time	Xyzzy	Science & Technology	26	2008-01-19 03:28
P3 TF time	PrimeCruncher	Software	30	2003-12-21 05:26

2016-01-03, 04:40	#2
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 16065₈ Posts	If it's single threaded, probably. On my i7-2600K, it's roughly 1-2 hours of ECM with 8 threads, before maybe ~12 hours of NFS with 8 threads. (These time estimates are order of magnitude.)

2016-01-03, 04:52	#3
jux Aug 2015 2·3³ Posts	I ran it with 8 threads though Old version of YAFU (1.24) never spent this long on ECM

2016-01-03, 08:01	#7
VBCurtis "Curtis" Feb 2005 Riverside, CA 1010100000111₂ Posts	the complete set of curves at 3M plus 541 curves at 11M won't take 2 hours with 8 threads, so it definitely wasn't running multi-threaded for the ECM phase. I wouldn't change the ratio- 541 curves @ 11M for that number sounds pretty good to me; I either last ran YAFU when the default ratio was higher, or I am simply mistaken.

2016-01-03, 08:45	#8
jux Aug 2015 2·3³ Posts	I have always considered ECM to be a quick check to remove some small factors - I don't want it running for more than an hour Any ideas on how to get it fully multithreaded?