What do you want to achieve?
A very simple neural network can learn load statistics
(in basic - which node is faster) and therefore it would be able to balance tasks more optimal than RoundRobin or Random policy.
I am working on it, and I am sure it'll work.
For example, if I have two nodes in cluster, where one is two times faster than other then RoundRobin isn't the best, is it.
However there are some minor practical problems, for instance, neural net will have to maintain the information it has learned (weights of neuron connections) somehow
so... practical meaning of it may not be very big.
But still interesting research topic.
BTW I have read that next version of WS is going to have
some kind of dynamic load balancing "based on server statistics".
I'll post my results in this forum.
This is a very interesting proposal - to do load balancing based on Neural Networks' concepts. Have you found any results? DO you need any help on this?
Smike, I can collaborate on this with you and find more stuff thro' reading.
> Have you found any results?
Yes, I have been busy writing my thesis.
But I am going to create a web page with overview
of algorithm and test results.
In short - dynamic load balancing provides up to two times better performance in heteregenous network.
I plan to publish the web page after a few days.
> DO you need any help on this?
It would be interesting to discuss it with you.
> Smike, I can collaborate on this with you and find
> more stuff thro' reading.
You can write me at email@example.com
> > Have you found any results?
You can look at the results here.
Dynamic load balancing algorithm gave much more performance than static load balancing in heterogeneous clusters.
As the result I got a master degree with distinction (comp. sc. major).