Cybertronic,

Can you elaborate on the following. How can we circumvent backtracking?

Thanks

[QUOTE=Cybertronic;172974]
BTW... we can splitt a TEST n in serveral tasks. Each core a 'RUN' ... and
there are exist more than one way to finished a certificate. It is possible to
circumvent the backtracking process...[/QUOTE]

[quote=Kosmaj;173115]
How can we circumvent backtracking?
[/quote]

First of all, it is a lot of manual work !

Backtracking is ever automatic in PRIMO, but we can compel a higher RUN from a TEST.
Example. PRIMO backtrack from TEST 6 RUN 1 index >100000 to TEST 5
RUN 4 , index= 4230, so we can manual compel TEST 6 RUN 2 i=1 or RUN 3,4,5 and 6 . Sometimes PRIMO find a point and jump to TEST 7 and the
backtrack to TEST 5 is unnecessary. (SOMETIMES!)
So we can calculate more than 1 way simultaneous.

For a overview ,all manual steps and changes come in a list, like Exel.

I mean, test the number 10^400+69 with primo. To begin, set in tmp-file manual for TEST "n" , ar=5 , ap=1, so you get also a valid certificate. Note , in Phase 2 it takes more time to finished this special point.

[quote=ET_;172970]That's where I came from! :smile:

When I try to compute .971057^(1/5) I have a different result from yours... So I was wondering where am I wrong?

Luigi[/quote]

[quote=philmoore;172972].863423 is .971057^5, however, this is the correct computation to make the estimate. (Norman says 4.86, and he has had as much experience as anyone using Primo on large numbers.) Luigi, change 1/5 to 5 or 4.86 and you will have a reasonable estimate. Of course, your estimate should get more accurate as you get a little further into the test.

Norman, you ran into trouble with Primo on a 10000+ digit number, didn't you? Was the problem that Primo would not run stage 2? I am guessing that Luigi's number will finish successfully, as it is not too much larger than the 7993 digit Primo record, and maybe even the next number on our list would work, but if we continue with Primo, we may well find that we have reached the breaking point eventually.[/quote]


Sorry about that Luigi. Phil is correct. Although the computation was correct, I mistyped. It should be the 5th power, not the 5th root like I typed.

[quote=Cybertronic;172968]Gary, how many cores do you use for the certificate ?[/quote]

Just one. I thought it was not possible to use more than one with the currently available Primo 3.0.6. I now see that it is possible but with some manual intervention using a modified version. If I can understand it, perhaps I'll put 2-3 cores on it after the end of May to get it finished up faster.


Gary

[quote=Cybertronic;173017]At this time , I play with "2^29727+20273" on 4 cores simultanious.

Version: PRIMO 3.0.6 - special[/quote]


I've now read the subsequent posts about doing this on more than one core and couldn't understand them. Are there any laymen's instructions for running on multiple cores? This is only about the 4th-5th time I've run Primo for something.

I'll be curious to see if it will work properly for you on an 8949-digit test.


Thanks,
Gary

[quote=gd_barnes;173137]I've now read the subsequent posts about doing this on more than one core and couldn't understand them. Are there any laymen's instructions for running on multiple cores? [/quote]

Okay, I will try it.

1. Save your last tmp-file under the PRIMO directory.
2. Create 4 PRIMO directorys. For each core one. Example: P1,P2,P3,P4
3. Copy all files in these four. (P1..P4)
4. Open the "master".tmp-file for edit.

You find : [X]

ve=3006
ce=4
pe=1
tx=0
sf=1
ar=4 -> RUNs , set here ar=1,2,3,4,5 or 6 , for each core another ar
ap=1..218000 -> actual index , for RUN 6 is it "delta" [ap=1..30000000]

The max index for the RUNs are:
1: about 100000
2: about 195000
3: about 220000
4: about 275000
5: about 140000

BTW , also you can splitt one RUN in example: ever ar=1
and then for the tasks ap=1, ap=10000, ap=50000, ap=80000

There are exist more intructions, but you will see for the first time , that is possible the finished a TEST faster or you can follow more then 1 way simultaneous.

Best Norman

Short info to PRIMO 3.0.6 - special

Marcel wrote to me on 20th October 2008 :
"
As promised here is a "special" 3.0.6 version. The sieving part is
done with 4 times more small primes. But as I already told you, do
not expect too much, 10000 digits is really difficult (in fact, it
requires other algorithms than the ones used by Primo).
"

Note, it is not a multicore version.

Cybertronic

Thank you for your detailed explanation. I'll try it on a small number (1000 digits or so) to test. However, avoiding backtracks can be complicated, I'm afraid, as they often appear in groups. How to decide which point to keep? Splitting a single run over multiple index intervals sounds easier.

And good luck wiht your large PRP, 29728 bits, almost 9000 digits!

Great! Thanks for the detailed explanation Cyber.

Like Kosmaj, I think I'll try it on a smaller PRP that would normally take Primo a few hours on one core and see if I can get it to work on 2 or 3 cores.

Note, it is not a child's play.

Is a task well advanced so stop the other tasks and save in a exel-file the last stand and save the tmp-file in another directory.
Example : TEST 10 RUN 2 index=30002, 23091 bit/ 23230 :
Copy tmp-file to *.10_23091.

That is important for a later manual backtrack, if the actual way a failure. E.g., there no other point to the next TEST found after TEST n RUN 6 , delta >30M.

Take the highest process tmp-file and overwrite the other tmp-files.Splitt
manual once more the tasks.

Later, for Phase 2 , we can also splitt on several cores.
Rewrite the line Type=-[1..4] to Type=[1..4] so PRIMO jump to the first
Type=-?

Example: core 1 take TEST 1 .. 10 , Type=- , core 2 take TEST 11..30,
so the first 10 you have to do "Type=n" , kill the "-"

3. Is Phase 2 done, copy in logical order all parts in a new tmp-file include the head-text of the original tmp-file.

4. You will see the signatur is invalid :-)
So you must verify new *.out-file per PRIMO
(Check signature only - make no "hook")

It takes 12h for a 6k number. After that , end of file we have a valid signature


Here a link for a demo picture:
[url]http://www.sendspace.com/file/r7t6uo[/url]

New picture:
Here a link for a demo picture with a [B][U]improved EXEL-table[/U].[/B]
[url]http://www.sendspace.com/file/8nb9rs[/url]