Daniel Quintiliani | 4 Sep 16:10 2004
Picon

Long-term projects: exponential and logarithmic growth


Remember a month ago I suggested that with long-term distributed computing projects,
logarithmic growth is a better estimate than exponential growth because people leave the
project or die as processing power grows?

I just looked at the OGRp1 graphs <http://www1.distributed.net/~pstadt/ogr/> and noticed 
that the processing power growth did not catch up to the amount of people who left the project.

However, when I plotted (this time by month, not half-month, because I didn't feel like it) estimates
of the rate in Tnodes/sec from every month of the OGRp1 project, the R^2 value of the logarithmic
growth was 0.0631 and the R^2 value of the exponential growth was 0.0453. Both values are low,
since it wasn't even close to logarithmic growth due to more people leaving than processor power 
growing, but logarithmic growth was still a better estimate if you added up all of the estimated average 
nodes/month (31,837,855,980,401,200,000 total for logarithmic, 24,832,936,897,178,500,000 total for
exponential) and the real average nodes/month (32,399,136,000,000,000,000 total). I don't know the actual
total OGRp1 stubs, does anyone know them or if I was close?

I also looked at the OGRp2 graphs and plotted by half-month. Even with only three month's data, logarithmic
growth still produced a better R^2 value (0.9246) than exponential growth (0.5271).

As for RC5 and those predictions, I updated the sheet at mid-August (452y 6mo R^2=0.8621) and September 1st (454y
6mo R^2=0.8645). So it basically comes down to the fact that RC5 will take at least 450 years to complete, not 17.

So, basically, logarithmic growth provides a better, but nowhere near perfect, estimate, and RC5 will
take AT LEAST
450 years to complete (from Dec 2002).

--

-Dan
(Continue reading)

Elektron | 5 Sep 09:08 2004
Picon

Re: Long-term projects: exponential and logarithmic growth


On 4 Sep, 2004, at 22:10, Daniel Quintiliani wrote:

> Remember a month ago I suggested that with long-term distributed 
> computing projects,
> logarithmic growth is a better estimate than exponential growth 
> because people leave the
> project or die as processing power grows?

Actually, people probably 'left' OGR because of a lack of packets, and 
now, because there's somehow more interest in RC5 (especially in 
coding/designing cores). Or perhaps it's because of other distributed 
projects.

As it stands, RC5-72 looks like it might be approaching some kind of 
growth. Of course, the RC5-64 keyrate looks like it was approximately 
stable for over a year (perhaps our computer power doesn't double every 
18 months). But RC5-72 is really an exercise in futility, because when 
people see that it'll take several hundred years, it just doesn't seem 
worthwhile.

That is, of course, why I run OGR. It's at least doable in a reasonable 
timeframe, though my computer (PPC 7450 ish) is much better at RC5 than 
OGR, relatively speaking.

We don't have the full history for RC5-64, but at least, people don't 
seem to have left, and people don't seem to be leaving RC5-72 either 
(like what appears to have happened with OGR).

Looking at the graphs, though, RC5-72 looks like it roughly 
(Continue reading)

Daniel Quintiliani | 6 Sep 14:52 2004
Picon

Re: Long-term projects: exponential and logarithmic growth


> On 4 Sep, 2004, at 22:10, Daniel Quintiliani wrote:
> 
> > Remember a month ago I suggested that with long-term distributed 
> > computing projects,
> > logarithmic growth is a better estimate than exponential growth 
> > because people leave the
> > project or die as processing power grows?
> 

> 
> Looking at the graphs, though, RC5-72 looks like it roughly 
> approximates some kind of linear growth (the dip in keyrate this summer 
> and last can probably be explained by people turning computers off 
> during the summer). 

A linear trendline produces:

y = 3.0401x + 53.134
R^2 = 0.8519
89 years 1 month

A logarithmic trendline produces:

y = 43.762Ln(x) - 3.6719
R^2 = 0.8645
454 years 6 months

They're both pretty close. However, I still think logarithmic growth is a better
estimate. Besides people leaving and dying, not everyone upgrades their
(Continue reading)

Elektron | 6 Sep 17:29 2004
Picon

Re: Long-term projects: exponential and logarithmic growth


On 6 Sep, 2004, at 20:52, Daniel Quintiliani wrote:

> They're both pretty close. However, I still think logarithmic growth 
> is a better
> estimate. Besides people leaving and dying, not everyone upgrades their
> computers right away, and due to compatibility problems, this may be
> worse in the future. Just look at the problems to get XP adopted, the 
> fact that
> no one is downloading XP SP2 because it breaks apps, and the coming
> Longhorn/TCPA/NGSCB, which will obsolete everything and require all new
> hardware.

Software updates are irrelevant. Hardware updates are necessary. Even 
poor lil' me has never had the same primary computer for more than 
about five years. And with computers always getting faster, a lot of 
people will have to leave to keep the growth logarithmic.

It's hard to explain the almost zero growth in the RC5-64 keyrate 
though. And our current keyrate still isn't up to par with the latest 
rc5-64 keyrates, either. It really takes a few more years' worth of 
data to say much, since the surge of interest in the first year or so 
were likely due to people who already knew about d.net.

People who don't like like "trusted computing" will still need to buy 
new computers at one point or another (though they may end up just 
buying parts and installing win2k). Or someone will reverse-engineer 
all the "trusted computing" stuff. And try getting companies in other 
countries to trust Microsoft. It just doesn't work.

(Continue reading)

waldo kitty | 6 Sep 21:58 2004
Picon

Re: Long-term projects: exponential and logarithmic growth


something else to consider in all this is that some machine default to round robin thru the contests... my
boxes were doing this at 
one time... ISR that i adjusted (some of) them to do OGR only so as to contribute to that project and get it out
of the way so that 
attention can be returned fully to RC5...

--

-- 
        _\/
       ( <at>  <at> )                      Waldo Kitty, Waldo's Place USA
__ooO_( )_Ooo_____________________ telnet://bbs.wpusa.dynip.com
_|_____|_____|_____|_____|_____|_____ http://www.wpusa.dynip.com
____|_____|_____|_____|_____|_____|_____ ftp://ftp.wpusa.dynip.com
_|_Eat_SPAM_to_email_me!_YUM!__|_____|_____ wkitty42 -at- alltel.net
_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5

waldo kitty | 6 Sep 22:05 2004
Picon

Re: Long-term projects: exponential and logarithmic growth

Daniel Quintiliani wrote:

> A linear trendline produces:
> 
> y = 3.0401x + 53.134
> R^2 = 0.8519
> 89 years 1 month
> 
> A logarithmic trendline produces:
> 
> y = 43.762Ln(x) - 3.6719
> R^2 = 0.8645
> 454 years 6 months
> 
> They're both pretty close. 

pretty close to what?? sure looks like a huge difference in 89 years and 454 years...

--

-- 
        _\/
       ( <at>  <at> )                      Waldo Kitty, Waldo's Place USA
__ooO_( )_Ooo_____________________ telnet://bbs.wpusa.dynip.com
_|_____|_____|_____|_____|_____|_____ http://www.wpusa.dynip.com
____|_____|_____|_____|_____|_____|_____ ftp://ftp.wpusa.dynip.com
_|_Eat_SPAM_to_email_me!_YUM!__|_____|_____ wkitty42 -at- alltel.net
_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5

(Continue reading)

Dave Huang | 6 Sep 23:05 2004

Re: Long-term projects: exponential and logarithmic growth

On Mon, Sep 06, 2004 at 04:05:35PM -0400, waldo kitty wrote:
> Daniel Quintiliani wrote:
> >A linear trendline produces:
> >
> >y = 3.0401x + 53.134
> >R^2 = 0.8519
> >89 years 1 month
> >
> >A logarithmic trendline produces:
> >
> >y = 43.762Ln(x) - 3.6719
> >R^2 = 0.8645
> >454 years 6 months
> >
> >They're both pretty close. 
> 
> pretty close to what?? sure looks like a huge difference in 89 years and 
> 454 years...

The fit is pretty close to the actual data points.
--

-- 
Name: Dave Huang         |  Mammal, mammal / their names are called /
INet: khym@... |  they raise a paw / the bat, the cat /
FurryMUCK: Dahan         |  dolphin and dog / koala bear and hog -- TMBG
Dahan: Hani G Y+C 28 Y++ L+++ W- C++ T++ A+ E+ S++ V++ F- Q+++ P+ B+ PA+ PL++
_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5

(Continue reading)

Dead J. Dona | 7 Sep 09:30 2004
Picon

RE: Long-term projects: exponential and logarithmic growth

> > They're both pretty close. 
> 
> pretty close to what?? sure looks like a huge difference in 
> 89 years and 454 years...

Huh huh..

Both numbers is bigger than my future life =)))))))))

So.... There's no difference....
_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5

Wacław Schiller | 7 Sep 12:42 2004
Picon

How to configure client to leave idle CPU time?

I want to configure ./dnetc not to use all of available CPU time.
I know it doesn't eat up all my CPU time if there is anything to do, the
problem is that one of the Windows machines I want the client to setup
at overheats if left at full usage for too long, and I'm looking for
a way to configure it nevertheless.
Torinthiel

--

-- 
 Waclaw "Torinthiel" Schiller       GG#: 542916
   torinthiel@...
   gpg: B06901F1 fpr: FAA3 559F CAE9 34DE CDC8  7346 2B6E 39F2 B069 01F1
 "No classmates may be used during this examination"
_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5

Elektron | 7 Sep 13:56 2004
Picon

Re: How to configure client to leave idle CPU time?


On 7 Sep, 2004, at 18:42, Wacław Schiller wrote:

> I want to configure ./dnetc not to use all of available CPU time.
> I know it doesn't eat up all my CPU time if there is anything to do, 
> the
> problem is that one of the Windows machines I want the client to setup
> at overheats if left at full usage for too long, and I'm looking for
> a way to configure it nevertheless.

Well, on the OSX client, there's a "Pause if processor temperature 
thresholds are exceeded" (in "general client options"), but I don't 
know if that also exists in Windows. It might be worth looking at, 
though.

- Purr

_______________________________________________
rc5 mailing list
rc5@...
http://lists.distributed.net/mailman/listinfo/rc5


Gmane