Ouch! Urgent help needed

karjalan · 2016-08-05, 20:56

Well, I just lost my reservation of an 100,000,000+ digit numbers exponent, which was 33% done. And of course I want it back. Please help me!

What I did: copied *.txt files from Windows Prime95 folder to Linux mprime folder, forgetting about copying output files. After running mprime the very first thing it did - it unreserved the biggest exponent. Holy shit! Absolutely disappointed, taking into account how many months of CPU time it took. Hope there is some way to get this reservation back (I still have all the needed files, only the reservation is missing).

I also propose to take that functionality away out of the program, I mean unreserving an exponent without any user interaction. At least ask user for (y)/(n) input.

Already tried to put N/A instead of reservation ID in worktodo.txt - doesn't work, after making manual communication with server this line disappears.

karjalan · 2016-08-05, 22:30

Fortunately not so urgent anymore, the main issue is solved, sorry for making panic.

Found help here:
http://www.mersenneforum.org/showpos...50&postcount=2
Got a new Assignment ID, everything seems to be ok. One additional step I had to do after acquiring new ID was replacing ",0" at the end of corresponding line with ",1", otherwise it started calculations with P-1 test, not with resuming LL test.

So I'm very happy now, but the second part of my message -- proposal to modify the software behavior -- is still relevant.

LaurV · 2016-08-06, 03:04

First, there are many reasons why mprime could want to unreserve your exponent, you switched not only the software, but also the hardware, it could consider the new hardware untrusted (no DC good result known), or too slow, etc. In fact, the main problem is that you didn't copy the checkpoint files: no work that is already started will be interrupted and unreserved. There is no need to modify anything on the server side/behavior here.

Second, if P95 wants to do P-1, it means there is not enough P-1 done for that exponent. I would, very politely, suggest that you do the P-1 first, or tell the exponent public and some big gun here around can do P-1 with a GPU, quite fast. That is because P-1 may find a factor, and save you CPU time, and you could do something else, more useful, with your CPU in that time... You may end up spending another month (or more) doing LL on an exponent where a small factor could have been easily found in few days or so of P-1.

Madpoo · 2016-08-06, 17:13

Quote:

Originally Posted by LaurV

...the main problem is that you didn't copy the checkpoint files: no work that is already started will be interrupted and unreserved.

Yes... that. If work is in progress (there's a save file present), it should not unassign anything. My guess... mprime started up without the old save file being present, and it unassigned.

But then as I think you discovered, putting the save file back there and adding the work line manually (with the old AID, or leaving the AID part of the line blank) it should create a new assignment for you.

Quote:

Second, if P95 wants to do P-1, it means there is not enough P-1 done for that exponent. I would, very politely, suggest that you do the P-1 first, or tell the exponent public and some big gun here around can do P-1 with a GPU, quite fast. That is because P-1 may find a factor, and save you CPU time, and you could do something else, more useful, with your CPU in that time... You may end up spending another month (or more) doing LL on an exponent where a small factor could have been easily found in few days or so of P-1.

Yeah... you probably should let it do P-1. You might find a factor and save yourself a LOT of effort.

I tried to look up the exponent in question and I probably got the wrong one because it's showing P-1 work as being done by you back in March. I also only see the one assignment (not an old expired one and a new one). So, I may have either the wrong user or the wrong exponent... (I was guessing your username based on your forum name here).

Whatever the case, as long as you have an assignment now and you're all good, then nothing to worry about.

karjalan · 2016-08-06, 20:53

Indeed, I do agree that it was purely my fault with forgetting to copy save files, and it caused all the thing. But still, "no work that is already started will be interrupted and unreserved" -- but in fact it was started, and P95 was theoretically able to figure it out even with save files missing. There was information on the server about 33% progress, perhaps in the same database table where assignment IDs are stored, so it could be checked without any additional query costs. Of course, if the save files are non-recoverably lost, there is no sense in treating exponent as "started" -- and it's exactly the way P95 went, instead of clarifying it from the user ("Do you still have the save files?"). Well, I do understand that it's pretty rare case, but if P95 could handle such a cases in a more user-friendly way, it would be only better.

I guess, it was just this 0 at the end of the line, what forced P95 to start a new P-1 test, wasn't it?
http://www.mersennewiki.org/index.php/Worktodo.txt
There is already one P-1 done for this exponent, so I think I had to reserve this number with 1 as the last parameter.

Madpoo, there was nothing wrong with your guessing. And about only one assignment -- I think it's because the old one was not expired but unreserved.

I repeated the same thing with a smaller exponent: reserving an exponent with P-1 done for a LL test, then unreserving, then reserving again manually with 0 as the last parameter -- and after finishing this second P-1 it said "Result was not needed". So I think there is no need to run P-1 test again for this bigger exponent too, and I tend towards just continuing LL test. It's noteworthy that both "secondary" P-1 tests (I mean both for the smaller and for the bigger exponent) are with slightly different (slightly lower) bound parameters (B1 and B2) compared to the "primary" P-1 tests which are already done.

I appreciate your answers, guys, I feel myself really supported.

2016-08-05, 20:56	#1
karjalan Aug 2016 3 Posts	Ouch! Urgent help needed Well, I just lost my reservation of an 100,000,000+ digit numbers exponent, which was 33% done. And of course I want it back. Please help me! What I did: copied *.txt files from Windows Prime95 folder to Linux mprime folder, forgetting about copying output files. After running mprime the very first thing it did - it unreserved the biggest exponent. Holy shit! Absolutely disappointed, taking into account how many months of CPU time it took. Hope there is some way to get this reservation back (I still have all the needed files, only the reservation is missing). I also propose to take that functionality away out of the program, I mean unreserving an exponent without any user interaction. At least ask user for (y)/(n) input. Already tried to put N/A instead of reservation ID in worktodo.txt - doesn't work, after making manual communication with server this line disappears.

2016-08-05, 22:30	#2
karjalan Aug 2016 3₁₆ Posts	Fortunately not so urgent anymore, the main issue is solved, sorry for making panic. Found help here: http://www.mersenneforum.org/showpos...50&postcount=2 Got a new Assignment ID, everything seems to be ok. One additional step I had to do after acquiring new ID was replacing ",0" at the end of corresponding line with ",1", otherwise it started calculations with P-1 test, not with resuming LL test. So I'm very happy now, but the second part of my message -- proposal to modify the software behavior -- is still relevant. Last fiddled with by karjalan on 2016-08-05 at 22:44

2016-08-06, 03:04	#3
LaurV Romulan Interpreter Jun 2011 Thailand 7×1,373 Posts	First, there are many reasons why mprime could want to unreserve your exponent, you switched not only the software, but also the hardware, it could consider the new hardware untrusted (no DC good result known), or too slow, etc. In fact, the main problem is that you didn't copy the checkpoint files: no work that is already started will be interrupted and unreserved. There is no need to modify anything on the server side/behavior here. Second, if P95 wants to do P-1, it means there is not enough P-1 done for that exponent. I would, very politely, suggest that you do the P-1 first, or tell the exponent public and some big gun here around can do P-1 with a GPU, quite fast. That is because P-1 may find a factor, and save you CPU time, and you could do something else, more useful, with your CPU in that time... You may end up spending another month (or more) doing LL on an exponent where a small factor could have been easily found in few days or so of P-1. Last fiddled with by LaurV on 2016-08-06 at 03:06

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
a rather serious (and urgent) problem...	ixfd64	Lounge	14	2008-11-19 12:25
[URGENT] Pain: troubles on implementing SIQS sieve	Hermes	Factoring	27	2008-10-14 13:54
URGENT: prime95 25.2 on Linux with elf64 :help:	T.Rex	Software	3	2007-07-04 14:29
Help needed	AntonVrba	Math	3	2007-03-06 10:55
V24.12 QA help needed	Prime95	Software	5	2005-06-17 15:54

2016-08-06, 20:53	#5
karjalan Aug 2016 3₁₀ Posts	Indeed, I do agree that it was purely my fault with forgetting to copy save files, and it caused all the thing. But still, "no work that is already started will be interrupted and unreserved" -- but in fact it was started, and P95 was theoretically able to figure it out even with save files missing. There was information on the server about 33% progress, perhaps in the same database table where assignment IDs are stored, so it could be checked without any additional query costs. Of course, if the save files are non-recoverably lost, there is no sense in treating exponent as "started" -- and it's exactly the way P95 went, instead of clarifying it from the user ("Do you still have the save files?"). Well, I do understand that it's pretty rare case, but if P95 could handle such a cases in a more user-friendly way, it would be only better. I guess, it was just this 0 at the end of the line, what forced P95 to start a new P-1 test, wasn't it? http://www.mersennewiki.org/index.php/Worktodo.txt There is already one P-1 done for this exponent, so I think I had to reserve this number with 1 as the last parameter. Madpoo, there was nothing wrong with your guessing. And about only one assignment -- I think it's because the old one was not expired but unreserved. I repeated the same thing with a smaller exponent: reserving an exponent with P-1 done for a LL test, then unreserving, then reserving again manually with 0 as the last parameter -- and after finishing this second P-1 it said "Result was not needed". So I think there is no need to run P-1 test again for this bigger exponent too, and I tend towards just continuing LL test. It's noteworthy that both "secondary" P-1 tests (I mean both for the smaller and for the bigger exponent) are with slightly different (slightly lower) bound parameters (B1 and B2) compared to the "primary" P-1 tests which are already done. I appreciate your answers, guys, I feel myself really supported.