[computer-go] Pay-as-you-go cluster?

[Mark Boon]

On 1-okt-06, at 19:13, David Doshay wrote:

> This will only work if you have an algorithm that is willing to  
> give up on
> nodes that are not reporting. Otherwise you cannot play an interactive
> game. In the @home computing model you never know when the owner
> will ask for their machine back, and at that time you loose all  
> predictability
> about when you will get your cycles again. It is not trivial to  
> distribute a
> number of jobs and then decide which of those jobs you can do without.
> In the SETI model it does not matter when the answer returns, but in a
> realtime game it does.

I did think a little bit about these issues of course. What you'll  
need is an iterative process where after a little computation time  
you have your best guess so far ready to play and then try to improve  
on it while you still have time. Tasks also need to be prioritized  
and important tasks should probably handed out to more than one CPU  
to improve the chance it gets completed in time. Once you have really  
large number of CPUs it's not a problem to have some redundancy.

The other thing to do is try to define tasks as such that the results  
can arrive 'late'. That means back-propagating the information gained  
and is not easy. I wouldn't be surprised that when these problems are  
being looked at and worked towards solutions (not just in Go, but in  
general) we also get a better understanding of how the brain works.

Mark

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://computer-go.org/pipermail/computer-go/attachments/20061002/da302262/attachment.htm