![]() |
![]() |
#1 |
May 2003
3×97 Posts |
![]()
I was thinking about pausing a factorization in msieve to finish a few high-priority jobs in prime95. If I stop gracefully with Control+c, will the program pick up where it left off when I restart it or is this likely to introduce errors into the data file?
|
![]() |
![]() |
![]() |
#2 |
"Curtis"
Feb 2005
Riverside, CA
10100110011012 Posts |
![]()
How to pause and restart depends on what phase of the job you're in.
Your thread title says "factmsieve", but your question says "msieve". Those are two different things. If you're in the sieving part of factmsieve, then your plan works just fine, and factmsieve resumes the job by restarting the current block of Q-ranges. If you're in the matrix-solving part of msieve, you'll want to restart msieve directly rather than through factmsieve, and you'll want to use -ncr flag to tell msieve "there is a matrix to resume, please resume it". Without -ncr, msieve will start the matrix over from 0 dimensions. I don't recall how to pass -ncr to factmsieve, which is why I call msieve directly in these circumstances (or I just don't cancel a factmsieve job when msieve is running the matrix). |
![]() |
![]() |
![]() |
#3 |
Jun 2012
DDB16 Posts |
![]() Code:
msieve -v -ncr -nc3 -t 4 The -nc3 tells msieve to move on the square root phase once LA is complete. If not, msieve will finish the linear algebra (-nc2) and then terminate without finding factors. You could then restart msieve again, this time with -nc3 but why bother? |
![]() |
![]() |
![]() |
#4 |
"Curtis"
Feb 2005
Riverside, CA
3×52×71 Posts |
![]()
The catch with restarting a matrix job that was created by factmsieve is that factmsieve has created file names different from msieve default.
I remember copying the msieve invocation line from factmsieve's screen display and adding in -ncr, so that msieve would find all the files. But it has been quite a few years since I've used factmsieve, so... you know. |
![]() |
![]() |
![]() |
#5 |
Jun 2012
67338 Posts |
![]()
And I have never used factmsieve, so take my comments from that perspective.
Thom - play with it a bit on smaller jobs is the best advice. Losing 4 hours of work is annoying - 2 months a disaster. |
![]() |
![]() |
![]() |
#6 |
May 2003
3·97 Posts |
![]()
As of this morning it was still sieving with about 70% of minimum relations. If I get back home and fid it doing matrix steps, i would just let it finish.
|
![]() |
![]() |
![]() |
#7 | |
Apr 2020
2×373 Posts |
![]() Quote:
![]() In all seriousness, if you're worried about things going wrong when you restart the script, make sure to back up the relations file (the big .dat file), and the .dat.cyc and .dat.mat files if they exist. Nothing should happen to these files, but it's better to be safe than sorry. |
|
![]() |
![]() |
![]() |
#8 | |
Jun 2012
1101110110112 Posts |
![]() Quote:
|
|
![]() |
![]() |
![]() |
#9 |
"Curtis"
Feb 2005
Riverside, CA
123158 Posts |
![]() |
![]() |
![]() |
![]() |
#10 |
Apr 2020
2·373 Posts |
![]() |
![]() |
![]() |
![]() |
#11 |
May 2003
3×97 Posts |
![]()
UPDATE:
Thanks for letting me know that it is safe to pause msieve. I found another workaround for the GIMPS jobs. Since the jobs are small, I simply opted to run them on another computer that has not been running any recent high-priority jobs. The first of these will finish in about 20 minutes. Now I am curious whether Primenet will take the job directly off the machine that was originally assigned the jobs, or whether they will disappear on the next completion estimate update, or I will have to manually remove them. |
![]() |
![]() |
![]() |
Thread Tools | |
![]() |
||||
Thread | Thread Starter | Forum | Replies | Last Post |
Stopping a primenet work fetch loop | SELROC | GPU Computing | 2 | 2019-03-31 10:40 |
Automatically stopping P-1 to benchmark? Why? | Uncwilly | Software | 2 | 2018-02-23 05:24 |
Stopping New Work | dmaine | Information & Answers | 1 | 2011-07-05 22:30 |
PrimeGrid PSP (Sieve) is stopping soon! | Joe O | Prime Sierpinski Project | 1 | 2010-11-04 23:05 |
Stopping on a dime? | R.D. Silverman | NFSNET Discussion | 1 | 2006-03-20 23:46 |