P-1 factoring Q&A thread - Page 3

Peter Nelson · 2005-05-22, 20:38

Allocating only half the RAM maximum on a dualcore appears wrong to me.

During usual test sequence TF , p-1, LL test, only the p-1 test (and only one stage of it) uses the lots of memory and this only runs for very small % of whole time (inc LL).

Therefore provided you have dualcore running 2 instances, it is very likely that both cores are not simultaneously doing P-1 stage.

Therefore the one doing the memory heavy work should be allowed to use more than half. When it completes and probably does LL then it will not use most of the memory therefore the other instance could then take it.

I therefore suggest that the maximum be not limited to half max mem but be configurable up to say 90% of total ram.

Joshua2 · 2005-05-23, 00:16

But, in the times when both are doing P-1 at once, there would be severe thrashing. Maybe the software could be made smart enough to "know" to make an exception in this case?
Also, if the two instances are staggered, about a month apart, maybe this would never happen since both cores are the same speed? However, when the computer is used for other tasks, it will probably always excute on core 1 first, and then if another high priority task is going on 1, then it would run on core 2. This would cause the p95 running on core 1 to gradually lose ground, and eventually, (maybe it would take years?) they would be having to do p-1 at the same time.
This problem should probably be solved, or efficiency will drop, especially in a couple years when more and more people get dual cores and processors.

dsouza123 · 2005-05-23, 13:19

The solution is getting more RAM.

At 1GB of RAM the desirable 185 MB for 33M
for each CPU/copy of Prime95 would take 370 MB
leaving 654 for the OS and all the other programs.

The idea is at least 512 MB per CPU core.
Which is in agreement with current single core recommendations
for a mid level PC.

It is a good idea to have plenty of RAM (not minimums),
it is an easy way to speed up your PC.
Virtual memory stays in RAM mostly, no thrashing due to paging swapping
memory onto the disk, RAM can cache repeatedly used files/programs
increasing speed.
With enough RAM the hard drive can power down,
saving power, reducing heat, reducing ware on the drive.

Remember even if 185 MB is allowed for Prime95, most of the time
(when not running P-1 stage 2 factoring) only about 12MB is used,
Prime95 doesn't allocate the RAM unless it needs it.

jasong · 2005-05-25, 20:44

From looking at the replies, I'd say people understand the problem and simply have different takes on it.

My main reason for concern, and this is the result of a thought experiment as opposed to an actual occurrence, is there are people who take Distributed Computing very seriously but have limited resources, so they don't necessarily want to supply the minimum amount of memory necessary for ALL the possible, but decide having enough for 95-99% is satisfactory.

It would be nice, but maybe too difficult for the programmers, to have a process that, instead of totally skipping p-1 when the other processor is running it, can "discover" the possibility that it can be run later, starts the other process, and goes back to p-1 when it's convenient. Also, what about the possibility that p-1 could stop BOTH processors in order to have the necessary memory for one p-1 instance I haven't analyzed this from a statistical standpoint, but it may be worthwhile to consider.

As I said, it may be too difficult for the programmers, but I thought it was at least worthy of discussion and appreciate people considering it.

Ding_Cong · 2005-06-11, 01:42

What meanings are these two Sentences?

"lgnoring suggested B1 value, using B1=385000 from the save file
lgnoring suggested B2 value, using B2=7988750 from the save file"

These two Sentences appear when I run zhe Prime95.exe .

please look at the picture:
http://www.equn.com/forum/attachment...ime95_KE27.gif

geoff · 2005-06-11, 05:32

When Prime95 begins a P-1 test it chooses the B1 and B2 parameters based on the memory available, and how long it expects an LL test to take (so it knows how much time would be saved if the P-1 test finds a factor).

If you are part way through a P-1 test and then change the memory settings, or install a new version of Prime95 (which might change the expected LL test time, or might have a different procedure for choosing parameters), it has to decide whether to start the test again with new parameters or to continue the current test with the old parameters.

The messages you see above are normal, they just say that Prime95 decided to continue with the old parameters instead of restarting the P-1 test.

2005-05-23, 00:16	#24
Joshua2 Sep 2004 13×41 Posts	But, in the times when both are doing P-1 at once, there would be severe thrashing. Maybe the software could be made smart enough to "know" to make an exception in this case? Also, if the two instances are staggered, about a month apart, maybe this would never happen since both cores are the same speed? However, when the computer is used for other tasks, it will probably always excute on core 1 first, and then if another high priority task is going on 1, then it would run on core 2. This would cause the p95 running on core 1 to gradually lose ground, and eventually, (maybe it would take years?) they would be having to do p-1 at the same time. This problem should probably be solved, or efficiency will drop, especially in a couple years when more and more people get dual cores and processors. Last fiddled with by Joshua2 on 2005-05-23 at 00:18

2005-06-11, 01:42	#27
Ding_Cong Oct 2004 Shanghai,China 2² Posts	What meanings are these two speeches? What meanings are these two Sentences? "lgnoring suggested B1 value, using B1=385000 from the save file lgnoring suggested B2 value, using B2=7988750 from the save file" These two Sentences appear when I run zhe Prime95.exe . please look at the picture: http://www.equn.com/forum/attachment...ime95_KE27.gif Last fiddled with by Ding_Cong on 2005-06-11 at 01:46

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Coordination thread for redoing P-1 factoring	ixfd64	Lone Mersenne Hunters	80	2021-01-19 17:17
NOT the official forum factoring project thread	jyb	Factoring	2	2013-09-03 16:11
Yet another basic-factoring-questions thread	davar55	Factoring	24	2011-01-23 23:57
Perpetual ECM factoring challenge thread...	Xyzzy	Factoring	65	2005-09-05 08:16
Deutscher Thread (german thread)	TauCeti	NFSNET Discussion	0	2003-12-11 22:12

2005-05-22, 20:38	#23
Peter Nelson Oct 2004 23² Posts	Allocating only half the RAM maximum on a dualcore appears wrong to me. During usual test sequence TF , p-1, LL test, only the p-1 test (and only one stage of it) uses the lots of memory and this only runs for very small % of whole time (inc LL). Therefore provided you have dualcore running 2 instances, it is very likely that both cores are not simultaneously doing P-1 stage. Therefore the one doing the memory heavy work should be allowed to use more than half. When it completes and probably does LL then it will not use most of the memory therefore the other instance could then take it. I therefore suggest that the maximum be not limited to half max mem but be configurable up to say 90% of total ram.

2005-05-23, 13:19	#25
dsouza123 Sep 2002 662₁₀ Posts	The solution is getting more RAM. At 1GB of RAM the desirable 185 MB for 33M for each CPU/copy of Prime95 would take 370 MB leaving 654 for the OS and all the other programs. The idea is at least 512 MB per CPU core. Which is in agreement with current single core recommendations for a mid level PC. It is a good idea to have plenty of RAM (not minimums), it is an easy way to speed up your PC. Virtual memory stays in RAM mostly, no thrashing due to paging swapping memory onto the disk, RAM can cache repeatedly used files/programs increasing speed. With enough RAM the hard drive can power down, saving power, reducing heat, reducing ware on the drive. Remember even if 185 MB is allowed for Prime95, most of the time (when not running P-1 stage 2 factoring) only about 12MB is used, Prime95 doesn't allocate the RAM unless it needs it.

2005-05-25, 20:44	#26
jasong "Jason Goatcher" Mar 2005 3505₁₀ Posts	From looking at the replies, I'd say people understand the problem and simply have different takes on it. My main reason for concern, and this is the result of a thought experiment as opposed to an actual occurrence, is there are people who take Distributed Computing very seriously but have limited resources, so they don't necessarily want to supply the minimum amount of memory necessary for ALL the possible, but decide having enough for 95-99% is satisfactory. It would be nice, but maybe too difficult for the programmers, to have a process that, instead of totally skipping p-1 when the other processor is running it, can "discover" the possibility that it can be run later, starts the other process, and goes back to p-1 when it's convenient. Also, what about the possibility that p-1 could stop BOTH processors in order to have the necessary memory for one p-1 instance I haven't analyzed this from a statistical standpoint, but it may be worthwhile to consider. As I said, it may be too difficult for the programmers, but I thought it was at least worthy of discussion and appreciate people considering it.

2005-06-11, 05:32	#28
geoff Mar 2003 New Zealand 13×89 Posts	When Prime95 begins a P-1 test it chooses the B1 and B2 parameters based on the memory available, and how long it expects an LL test to take (so it knows how much time would be saved if the P-1 test finds a factor). If you are part way through a P-1 test and then change the memory settings, or install a new version of Prime95 (which might change the expected LL test time, or might have a different procedure for choosing parameters), it has to decide whether to start the test again with new parameters or to continue the current test with the old parameters. The messages you see above are normal, they just say that Prime95 decided to continue with the old parameters instead of restarting the P-1 test.