P-1 Question - Page 4

Dubslow · 2012-01-09, 06:31

Quote:

Originally Posted by axn

P-1 will stop being short of resources when P95 default of 8MB is changed into something more meaningful (say 300MB or so). That 8MB figure is just way out of date.

I also agree, and it is in that spirit that I originally asked my question, though of course I'm trying to treat the symptoms, not the problem, when we're already short of resources.

cheesehead · 2012-01-09, 10:14

Quote:

Originally Posted by Dubslow

Quote:

Originally Posted by axn

P-1 will stop being short of resources when P95 default of 8MB is changed into something more meaningful (say 300MB or so). That 8MB figure is just way out of date.

I also agree, and it is in that spirit that I originally asked my question, though of course I'm trying to treat the symptoms, not the problem, when we're already short of resources.

Let's review why the default available memory figure is 8MB.

George Woltman has repeatedly stated that in order for prime95 to be successfully used on large numbers of systems, it needs to have as little impact on its host system operation as possible in its default configuration. We're going to get deployment into large arrays of school systems and such _only_ if the admins of such systems don't get complaints about how prime95 affects use of those systems for their official uses.

That's why the default value is only 8MB. It's NOT "way out of date". It is deliberately and purposely set low so as not to cause page swapping to slow down its host system if the host is heavily loaded with its intended tasks. This default value should NOT be changed.

Again, our choice is whether:

(1) we have prime95 as a guest on a system, doing stage 1-only P-1, or

(2) we get prime95 kicked off a system because its memory demands interfere with the other tasks for which that system is really intended.

Now, in a particular case, if our GIMPS representative can get agreement with the administrators to allow prime95 to allocate 300MB or so for stage 2 P-1, then it's up to our representative to set that parameter in the control files for those systems.

Those of you who are frustrated by the thought of the skimpy defaults in use on so many systems need to be reminded of the realities with which we work. Some of our GIMPS users are able to get prime95 installed and running on hundreds of systems -- but not by pissing off non-GIMPS users and administrators of those systems, which, after all, were bought and installed by someone else for some purpose other than GIMPS. You need to be just a bit grateful that some GIMPS folks are able to get massive numbers of institutional systems working for us, and not insist on killing the geese that lay the golden eggs.

cheesehead · 2012-01-09, 10:24

Quote:

Originally Posted by Dubslow

when we're already short of resources.

"We" are short of resources?

GIMPS is basically in the position of politely begging people who have systems that were bought to use for other purposes to let us slip in a special-purpose program that doesn't cause any trouble.

It isn't helpful for GIMPSters to start exhibiting an attitude that we are somehow entitled to more resources on those systems (that are not ours) on which we have voluntarily been given the use of background time.

- - -

Dubslow, you were probably thinking only of the balance of various work within GIMPS, but we all need to keep in mind just where we are in relationship to the donors of system time, and become aware of when our proposals to change default values in GIMPS will cause us trouble in our external relationships, which might easily lead to our losing those external donated resources and thus slowing down GIMPS progress.

cheesehead · 2012-01-09, 10:35

Quote:

Originally Posted by chalsall

cheesehead... based on your superior knowledge of P-1 work (at least, superior to mine), what do you think should be the parameters for the (possibly) new work type "P-1 Stage 1 and 2"? I assume it would be a function of the TF level and B1 done (where B1 == B2), but I don't understand the characteristics well enough to design an intelligent assignment algorithm.

I'll get back to this later -- not enough time right now. If you don't hear from me on this by next weekend, please send me a reminder.

NBtarheel_33 · 2012-01-09, 11:16

Quote:

Originally Posted by cheesehead

That's why the default value is only 8MB. It's NOT "way out of date". It is deliberately and purposely set low so as not to cause page swapping to slow down its host system if the host is heavily loaded with its intended tasks. This default value should NOT be changed.

Question: What was the minimum typical RAM, call it x, in a computer at the time that the 8 MB threshold was chosen? What is the minimum typical RAM, call it x*, in a computer now? Would it not be fair to claim that we ought to be able to adjust the default P-1 RAM to a value closer to z, where x*/z = x/8? Or, if you want to be extra careful, how about taking z to be half of the value that is obtained from the equation x*/z = x/8?

We have to be careful not to get in our donors' way, yes, but perhaps there was a time when we risked this more so by blithely asking for 8 MB. Maybe asking for 8 MB back in 1997 is like asking for 128 MB today.

axn · 2012-01-09, 11:37

EDIT- I wrote a reply to cheesehead, but NBtarheel_33 pretty much covered it :)

axn · 2012-01-09, 11:44

Quote:

Originally Posted by cheesehead

"We" are short of resources?

I took that as "we" (the people who have responded to George's call for P-1 volunteers) "are short of resources" (not enough of us to keep up with the LL).

Not very charitable of you to interpret things in the worst way possible.

Brian-E · 2012-01-09, 12:11

Quote:

Originally Posted by NBtarheel_33

Question: What was the minimum typical RAM, call it x, in a computer at the time that the 8 MB threshold was chosen? What is the minimum typical RAM, call it x*, in a computer now? Would it not be fair to claim that we ought to be able to adjust the default P-1 RAM to a value closer to z, where x*/z = x/8? Or, if you want to be extra careful, how about taking z to be half of the value that is obtained from the equation x*/z = x/8?

We have to be careful not to get in our donors' way, yes, but perhaps there was a time when we risked this more so by blithely asking for 8 MB. Maybe asking for 8 MB back in 1997 is like asking for 128 MB today.

I imagine that 8 MB was considered the minimum amount of memory required by the client software to run an LL test and/or to run at all. That value has probably hardly changed in the intervening years. Your point about memory being cheaper and more easily available now is well made, but the principle that extra memory over and above absolute minimum requirements should be specifically configured by the user is paramount, for the reasons which cheesehead discusses.

Further, if your suggestion of 128MB being equivalent to 8MB in the early days of GIMPS is in the right ballpark, then we're not out of the woods anyway. 128MB is inadequate for P-1 stage 2. So why compromise the principle at all?

axn · 2012-01-09, 13:30

Quote:

Originally Posted by Brian-E

I imagine that 8 MB was considered the minimum amount of memory required by the client software to run an LL test and/or to run at all. That value has probably hardly changed in the intervening years.

The 8MB figure is strictly _additional_ memory for P-1 (and ECM) stage 2. It doesn't reflect whatever memory P95 needs to run TF, LL or P-1 stage 1. P95 uses _whatever_ memory it needs to run these things -- the 8MB setting has no bearing on it. [A 3072K FFT is 24MB in size].

I love it when people read too much "principle" into pragmatic decisions.

Dubslow · 2012-01-09, 19:55

I am pretty sure that x*>1GB today -- though we must consider that in many schools, computers aren't up to date, so they're very probably more from around 2008 plus or minus a few years. Still, I would think that x*>=1GB.

New computers in ~2005 had 512 MB typically. (We have a couple of laptops and an old desktop in the house from then.)

cheesehead · 2012-01-09, 23:07

Quote:

Originally Posted by axn

I took that as "we" (the people who have responded to George's call for P-1 volunteers) "are short of resources" (not enough of us to keep up with the LL).

So did I -- except that I distinguish between two distinct types of "resource": (1) human volunteer administrators of computer systems and (2) the hardware/software resources of the computer systems administered by the human volunteers.

In other words, "we" (the human P-1 volunteers) have not enough volunteered systems with enough memory so that a significant amount can be dedicated to P-1, to keep up with the LL.

"We" (the human P-1 volunteers) can each administer dozens or hundreds of systems running P-1 -- so it's not a matter of a shortage of human resources, but it is a matter of insufficiency of the computer resources that "we" the human volunteers can provide for the stage 2 P-1 effort. "We" don't collectively have enough of such systems useful for stage 2 P-1 to keep up with the LL.

Quote:

Not very charitable of you to interpret things in the worst way possible.

No, it wasn't very charitable of you to interpret my response in your unfavorable way without asking for clarification, was it?

2012-01-09, 11:37	#39
axn Jun 2003 5,051 Posts	EDIT- I wrote a reply to cheesehead, but NBtarheel_33 pretty much covered it :) Last fiddled with by axn on 2012-01-09 at 11:39

2012-01-09, 19:55	#43
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 3×29×83 Posts	I am pretty sure that x>1GB today -- though we must consider that in many schools, computers aren't up to date, so they're very probably more from around 2008 plus or minus a few years. Still, I would think that x>=1GB. New computers in ~2005 had 512 MB typically. (We have a couple of laptops and an old desktop in the house from then.)