On 8 Jun 2001, at 19:50, Guido Lorenzini wrote:
> Don't like to take up your time, but the behaviour of Prime95 on a
> 750@795 Duron, P-1 factoring on 33.3mio exponent (B1=350000 and
> B2=4025000), is quite strange. The Duron completed the stage 1 without
> any problem, but during the stage 2 the time per iteration has slow
> down a lot, from 0.650 sec/it to many minutes for just one iteration.
With much disk thrashing? Sounds like your system is paging heavily,
which means that you have set the P-1 factoring memory too high.
How big you can get away with depends on the operating system but you
should definitely not allow P-1 more than 32 MB less than the total
memory on the system. 48 MB is safer especially with Windows NT/2000.
Note that the P-1 factoring memory limit applies in stage 2. Stage 1
only uses about as much memory as LL testing.
You've probably just come across this because of the exponent size.
Smaller exponents will use less memory even in stage 2.
> Above all the pc is unusable, so I've decided to edit manually
> worktodo.ini: from Test=33306743,68,0 to Test=33306743,68,1 to
> force prime95 to skip the factoring stage when the process was done
> for about 30% (The intermediate file mX306743 sizes 8,132 KB). I've
> tried to put everything on another machine but with no results: the
> behaviour was the same. Skipping the P-1 factoring stage shouldn't
> invalidate the LL test, as George says in undoc.txt:
>
> You can force prime95 to skip the trial factoring step prior to
> running a Lucas-Lehmer test. In prime.ini add this line:
> SkipTrialFactoring=1
>
> By the way: I've done it but apparently nothing has changed. If I've
> understood properly, adding this line is like to manually edit
> worktodo.ini. Isn't it?
Sensible. But you probably need to exit Prime95 & restart it to get
it to take any notice of a change to prime.ini or local.ini. Stop &
continue should suffice if you're just changing worktodo.ini. Anyhow
I don't know what will happen if you try to suppress P-1 factoring
halfway through the run!
>
> Is it better if I force prime95 to start p-1 factoring from zero once
> again, hoping any problem occurs this time? What may be the problem?
> Has anybody ever observed something like this?
It's probably not worth the effort. Changing the ,0 to ,1 at the end
of the worktodo.ini line tells Prime95 that P-1 has been completed.
That should fix your problem. But change the P-1 memory before the
next run ... e.g. if you have 128 MB RAM set both DayMemory &
NightMemory in local.ini to 80 or less.
I had a mild case of page thrashing in P-1 stage 2 on exponent
40250087 with a system running Win 2000 with DayMemory=96,
NightMemory=96 on a 128 MB system despite a total lack of other load
on the system. I recovered by restoring a backup of the Mxxxxxxx
savefile taken the last day that stage 1 was running, changing
DayMemory & NightMemory to 88 and restarting Prime95. Even that was
close to the bone as page thrashing would set in as soon as anything
else started running. Fortunately in my case this wasn't often, I
wasn't relying on the system to do anything else.
>
> Last question: is it normal that prime95 runs a self test each time it
> is going to test with different FFT sizes?
Yes. It will do this before starting P-1 factoring. The code executed
during P-1 stage 1 is _very_ similar to LL testing, so doing the
selftest at that point makes sense.
The next time it wants to use the same FFT size, it will remember
that the selftest has already done. The FFT sizes which have been
selftested are noted in local.ini. If you change the system in a
major way (upgraded CPU, more memory etc.) it probably makes sense to
remove the SelfTest lines from local.ini to force it to repeat the
selftests.
Regards
Brian Beesley
_________________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
Mersenne Prime FAQ -- http://www.tasam.com/~lrwiman/FAQ-mers
