Reserved for MF - Sequence 3408

[VBCurtis]

168 hr later, I have factors! Second dependency, else it would have completed under 7 days; as it is, I needed the help from Daylight Savings to make it under 7 days. :)

Some stats:
143.4M raw relations w/31-bit LP, 123.4M unique
147 hr sieving on 6 HT cores (900 CPU-hr, or 1800 thread-hr)
5.9M matrix at target density 90, 20 hr to solve on 6 threads (linux doesn't manage HT well- some cores were idle during this time; I will experiment with more threads to see if lower idle-core-time makes up for the thread contention)

The sequence rises through i1595; a c164 is in the middle of a t40 presently. I'll update factordb later today.

[VBCurtis]

i1591 through 1594 are added to factordb. 1595 is listed as c175, but aliqueit/ecm is currently working on a c164 (I think that since I invoked aliqueit to update the sequence to factorDB while another copy was still running, it didn't report the trivial prime found).

The c164 has had over half a t45, with the second half finishing overnight. Any public efforts (Sean?) should begin at t50 level.

The sequence has grown to 189 digits.

[swellman]

Ok I'll start ECM at t50 on the C164 later today. How much ECM do you want before starting GNFS?

[fivemack]

Originally Posted by VBCurtis

5.9M matrix at target density 90, 20 hr to solve on 6 threads (linux doesn't manage HT well- some cores were idle during this time; I will experiment with more threads to see if lower idle-core-time makes up for the thread contention)

I get round the Linux scheduling problem with 'taskset -c 0-5 msieve ... ... -t 6' to lock it to running on one thread per core

[Xyzzy]

Originally Posted by fivemack

I get round the Linux scheduling problem with 'taskset -c 0-5 msieve ... ... -t 6' to lock it to running on one thread per core

We just recently started doing this at fivemack's suggestion and it works wonderfully!



We actually modify the process after it starts, but that accomplishes the same thing and uses the same 'taskset' program.

taskset -p -c x y

where x is the core number and y is the process id.

[VBCurtis]

Thanks for the protip!! I still use the python script; can I call taskset from the script by adding the command ahead of factmsieve's invocation of msieve {...} -nc2?

[VBCurtis]

Originally Posted by swellman

Ok I'll start ECM at t50 on the C164 later today. How much ECM do you want before starting GNFS?

I use 0.31 * digits for GNFS, which gives me about 1.5t50 here. A C164 is a nontrivial factorization, but one I'd be likely to try in December if it is deemed too small for NFS @ Home and nobody else picks it up.

The t45 finished overnight.

[swellman]

I should finish a single t50 in a few days, and willing to run it up to 5000 curves at t55 level as well. Beyond that it belongs to GNFS.

[Xyzzy]

Originally Posted by VBCurtis

Thanks for the protip!! I still use the python script; can I call taskset from the script by adding the command ahead of factmsieve's invocation of msieve {...} -nc2?

Yep!

https://www.google.com/search?q=site...taskset+msieve

[swellman]

The C164 completed t50 with no hits. Running another 5k curves @B1=110M. Should be done by Sunday.

[swellman]

5000 curves completed @B1=110M, with no hits. Releasing the C164.