Neural Network Scheduler

[Robert Garrett]

"Denis Lagno" <dlagno at xxxxxxxxxx> wrote in message 
news:20050717172509.GJ549 at xxxxxxxxxxxx
> On Sat, Jul 16, 2005 at 01:25:29PM +0200, Sascha Retzki scribed:
>>
>> What do you think about a Neural Network Scheduler as described in [1] 
>> and
>> implemented in [2]?
>>
>> I like the idea, but did not have the time to test it yet. And honestly,
>> I know basicly nothing about scheds or neural networks ;-) ,
>> but found it to be interesting. At least worth the look.
>>
>> [1]: http://prdownloads.sourceforge.net/nnsched/nnsched.pdf?download
>> [2]: http://nnsched.sf.net/
>
> I've just superficially read the paper.
> AFAIU this scheduler requires a lot of changes to userland.
> For example X Server must report "features" about connected processes to 
> features collector device.
>
> Then, I do not quite understand how features would be propagated to 
> dependant processes.
> For example if you listen music via process1 | process2 > /dev/audio
> Features of process2 must be somehow propagated to process1
>
> It will be quite messy stuff..  At least OpenBSD guys would not commit 
> it:)

with some thought the kernel could handle the features part dynamically.. 
this is a interesting idea maybe worth playing with.

not certain how practical it will be though.

rob