mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > PrimeNet

Reply
 
Thread Tools
Old 2011-10-26, 03:00   #749
Christenson
 
Christenson's Avatar
 
Dec 2010
Monticello

6B016 Posts
Default

Quote:
Originally Posted by James Heinrich View Post
Congrats nucleon, that's quite the achievement! (You even beat out post #711 on the previous page of this thread )

Back to topic, I understand where Craig was going with his idea, but it's probably best to achieve the desired effect with more tweaking to the "Whatever makes sense" PrimeNet worktype. If L-Ls were only handed out to machines that can't reasonably run P-1 (due to lack of RAM) and anyone cable of running it was assigned P-1, then the assignment landscape would change very quickly.
I'll throw one at you: I have a 4-core Phenom II x4, and I just brought it on-line for GIMPS. It proceeded to start 4 LL tests BEFORE I could configure it to do P-1, and tell it it could use a Gigabyte or two of RAM with P-1. So I got 4 preliminary P-1s, and 4 LL tests got started, and I didn't feel like fixing it up, so they will complete, before it starts turning in a bit more P-1.

Is there a way to detect multiGigabyte machines at installation/startup and ask if the required 300M of RAM can be used for P-1 testing, before the LL tests get started?
Christenson is offline   Reply With Quote
Old 2011-10-26, 03:22   #750
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2×3×52×67 Posts
Default

I think I have the components to put together for an Opteron dual-core, 2.4GHz, with 3GB of RAM. No real GPU ability, I think. It's an ATI All in Wonder 9800 Pro, and the motherboard is only AGP. But this would work pretty well as a P-1 machine? Something else?
kladner is offline   Reply With Quote
Old 2011-10-26, 04:59   #751
axn
 
axn's Avatar
 
Jun 2003

7×683 Posts
Default

Quote:
Originally Posted by fivemack View Post
But optimising one is the same as optimising the other; you save GHz-days only by finding a factor, and the number of GHz-days you save (for the two LL tests) is not a function of the effort required by the P-1.

(a 54M takes about 180 GHz-days for two LL tests, so P-1 testing that costs less than 180 GHz-days per expected factor is worth doing; but parameters that cost 65 GHz-days per expected factor should definitely be used ahead of parameters that cost 90 GHz-days per expected factor)
Not quite. Consider the hypothetical case below:
Code:
LL      P-1   Prob   Effort Effort   Factors/
Cost    Cost  Succ   saved  spent    Unit time
----------------------------------------------
100     1.0   0.10    9.0   91.0     0.100
100     1.5   0.12   10.5   89.5     0.080
100     2.0   0.13   11.0   89.0     0.065
Highest factors/unit time is option 1. Most profitable for project is Option 3. And this is the basis of calculation today.

However, in the context of this thread, this is not the correct calculation. It is not a local (i.e. per exponent) optimization problem, but a global optimization problem. i.e Given the amount of compute power we have for dedicated P-1, what is the most number of factors we can find, while staying ahead of the LL wave. Naturally, this is a dynamic point which will change based on the ratio of compute power available for LL vs compute power available for P-1. So it might turn out that option 2 is the right one in the given context.

Anyway, my gut feel is that B/S extension can safely be ditched in this context.
axn is offline   Reply With Quote
Old 2011-10-26, 10:42   #752
Mr. P-1
 
Mr. P-1's Avatar
 
Jun 2003

7×167 Posts
Default

Quote:
Originally Posted by axn View Post
However, in the context of this thread, this is not the correct calculation.
To make your point more explicitly, it is the correct calculation of a client which is doing a P-1 prior to LLing the same exponent. It is not the correct calculation for a client which takes P-1 assignments specifically.

Quote:
It is not a local (i.e. per exponent) optimization problem, but a global optimization problem.
Yes. This is a point I have made repeatedly over the months.

Quote:
i.e Given the amount of compute power we have for dedicated P-1, what is the most number of factors we can find, while staying ahead of the LL wave.
No, it is not the number of factors which we find, which needs to be maximised, but the GHz-Days saved on other machines, by us doing the P-1 calculation, per GHz day spent by us. This depends, not just upon the P-1 we do, but upon the P-1 those other machines would do if we don't.

So using your hypothetical figures, in the same time, we could do six of option 1, four of option 2, or three of option 3. Suppose the average LLing machine does an option 1 P-1 on every non-P-1ed exponent it gets. If we did 6 option 1s, we would save just the P-1s on the LLing machines, for a ratio of work saved to work done by us of 1:1.

Or we could do 4 option 2s. There is a 10% chance that we find a factor that the LLing machines Option 1 would also find, saving only that LLing machines Option 1 P-1. There is also a 2% chance of finding a factor that the Option 1 machine would not have found, saving an Option 1 + LL cost. There is a 88% chance of finding no factor, again, only saving the LL machines' option 1 P-1. combining these probabilities, we will certainly save the LL machines' option 1 P-1 and we have a 2% chance of saving the LL cost which would not otherwise have been saved. The total saving is 4 * (1 + 0.02 * 100) = 12 the cost to us is 6 giving us a ratio of 2:1

Or we could do 3 option 3s. Using similar reasoning the total saving is 3 * (1 + 0.03 * 100) = 12, also giving us a ratio of 2:1.

In this example, options 2 and 3 are equally good, but if option 1 was slightly more expensive, that would tip the balance in favour of us doing option 2, while a cheaper option 1 would tip the balance the other way.

Quote:
Naturally, this is a dynamic point which will change based on the ratio of compute power available for LL vs compute power available for P-1.
That's another consideration. The above analysis assumes that we really can do 4 Option 2s on exponents to be LL tested in the near future. If in fact, there are only three such available, and our fourth test just pushes the P-1 wavefront out further and further ahead of the LLs, then Option 3 would be more desirable.
Mr. P-1 is offline   Reply With Quote
Old 2011-10-27, 02:58   #753
Christenson
 
Christenson's Avatar
 
Dec 2010
Monticello

24·107 Posts
Default

Quote:
Originally Posted by kladner View Post
I think I have the components to put together for an Opteron dual-core, 2.4GHz, with 3GB of RAM. No real GPU ability, I think. It's an ATI All in Wonder 9800 Pro, and the motherboard is only AGP. But this would work pretty well as a P-1 machine? Something else?
P-1 with 2 cores and 2 Gigs of RAM?? Yep....just get it an Xubuntu disk, so the OS doesn't eat it alive.
Christenson is offline   Reply With Quote
Old 2011-10-27, 03:56   #754
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2·3·52·67 Posts
Default

Quote:
Originally Posted by Christenson View Post
P-1 with 2 cores and 2 Gigs of RAM?? Yep....just get it an Xubuntu disk, so the OS doesn't eat it alive.
(Aside from the fact that there are 3 Gigs of RAM.) Yes. I was thinking that I would do a Linux setup if I get the hardware together. The alternative would be XP-32, and I have daily examples of how slowly things run there, compared with 64bit. I switch back and forth between XP-32 and Win7-64 every day and the difference, especially in mfaktc is glaring. I haven't made the same comparisons in Prime95, but I suspect that the situation is similar.

This just involves deconstructing an original Athlon box and putting the Opteron motherboard and PSU into the case. It will only have IDE drives, but I think that is a minor issue as none of the Prime apps are HDD constrained. It won't matter otherwise as this will be a dedicated box.

Does VPN work on Linux? That would be the easiest way for me to control such a system: over my network.

Last fiddled with by kladner on 2011-10-27 at 03:59
kladner is offline   Reply With Quote
Old 2011-10-27, 04:36   #755
delta_t
 
delta_t's Avatar
 
Nov 2002
Anchorage, AK

3·7·17 Posts
Default

Quote:
Originally Posted by kladner View Post
Does VPN work on Linux? That would be the easiest way for me to control such a system: over my network.
SSH won't work?
delta_t is offline   Reply With Quote
Old 2011-10-27, 04:38   #756
Dubslow
Basketry That Evening!
 
Dubslow's Avatar
 
"Bunslow the Bold"
Jun 2011
40<A<43 -89<O<-88

3·2,399 Posts
Default

Quote:
Originally Posted by delta_t View Post
SSH won't work?
lol

Why do you still use XP-32 at all?
Dubslow is offline   Reply With Quote
Old 2011-10-27, 04:41   #757
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2×3×52×67 Posts
Default

Quote:
Originally Posted by delta_t View Post
SSH won't work?
Ignorance in play at this end. I only need something I can set up to control over the LAN from a Windows box to Linux. VPN is what I know for Windows to Windows.
kladner is offline   Reply With Quote
Old 2011-10-27, 04:42   #758
delta_t
 
delta_t's Avatar
 
Nov 2002
Anchorage, AK

3·7·17 Posts
Default

Quote:
Originally Posted by Dubslow View Post
lol

Why do you still use XP-32 at all?
How is SSH XP-32? It's how I access my Unix servers.
Ahh, the simultaneous posting at work again. There are some Xp-32 SSH freeware clients.

Last fiddled with by delta_t on 2011-10-27 at 04:45
delta_t is offline   Reply With Quote
Old 2011-10-27, 04:43   #759
kladner
 
kladner's Avatar
 
"Kieren"
Jul 2011
In My Own Galaxy!

2·3·52·67 Posts
Default

Quote:
Originally Posted by Dubslow View Post
lol

Why do you still use XP-32 at all?
I have legacy hardware and software that doesn't like Win7 (Nikon film scanner, for one).
kladner is offline   Reply With Quote
Reply

Thread Tools


All times are UTC. The time now is 23:31.

Sun Nov 29 23:31:51 UTC 2020 up 80 days, 20:42, 3 users, load averages: 0.97, 1.09, 1.23

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.