[Beowulf] why we need cheap, open learning clusters

[Vincent Diepeveen]

This is why i have my own cluster at home!

Note that sensitive software you can never boinc.

Game software in general you don't want to boinc.
Only for open source software you can do something like that.

Note that the open source chess software that's there, that the  
parameter tuning also gets done at closed doors.
The way how the parameter tuning happens is very secret, also for the  
open source software.

In fact even the code that can modify the parameters of the open  
source chessprograms is secret.
So the code htey spread of those programs doesn't have the code that  
can modify the parameters.

Really 100% of all information how and what and where the parameter  
tuning takes place gets kept secret.

Some (ex-)NASA/NCSA guys are very busy there though in those open  
source projects.

Now from an algorithmic viewpoint those chess engines are not so  
interesting of these guys,
just the way how the parameter tuning works is interesting,
yet if you ask online they just spread provable desinformation about  
what happens there.

Having an oracle that chess technical information to which you can  
tune the parameters to, is very important to have.

For *some* of the projects it seems they somehow use the same oracle,  
which is very weird if you think about it.

Creating that oracle requires massive supercomputers...

On May 13, 2013, at 8:09 AM, Jonathan Aquilina wrote:

>
>
>
> On Mon, May 13, 2013 at 8:04 AM, Vincent Diepeveen <diep at xs4all.nl>  
> wrote:
>
> On May 12, 2013, at 7:55 PM, Lux, Jim (337C) wrote:
>
> >
> >
> > I just ran across an interesting anecdote (in Malcolm Gladwell's
> > "Outliers").  It's in the context of Bill Joy, who commented that
> > using timesharing and interactive systems compared to traditional
> > batch/card deck submission was like speed chess vs chess by mail.
> > That interactivity facilitated his spending thousands of hours
> > working with software.
>
> > [snip]
>
> > I am a BIG believer in personal computing…
> >
>
> It's funny that you mention chess and personal computing at the same
> time.
>
> Now i ran in 2003 Diep at a supercomputer. Diep is a chessprogram.
>
> With respect to the previous mail i wrote regarding the university
> Utrecht, until recently neary every year in the top50 universities of
> this planet,
> it's interesting to mention that at university, after a while i was
> allowed to run my chessprogram.
>
> Yet other students when they tried to launch my chessprogram, it got
> killed.
> Official reason given: it was eating too much RAM.
>
> In fact it was eating 8 MB ram. All machines had 64MB ram or more.
> Reason to eat 8 MB is that i had discovered this to run a lot faster
> at the unix machines (partly HP 60Mhz).
>
> Not sure whether this had to do with a caching issue of the processor
> or other circumstances.
>
> Yet i argue that centralized supercomputing is really slow way to
> develop your software. Personal computing is simply a faster way to
> develop your codes.
>
> The real disadvantage of supercomputers is that you always have to
> wait for weeks if not months for a batch of a few hours to get  
> executed.
> In that sense supercomputing is a lot slower than chess by mail,
> where you have 1 day a move on average.
>
> Vincent then why not develop your project for BOINC at least its  
> distributed computing, yes youll still have to wait for the results  
> but its getting crunched on either a graphics card or cpu depends  
> on how you code it.