Strategy Optimization > Rebalance work when agents finish before others

New comment
 
Hello everyone!

First of all, forgive me if this topic is repeated, but I did not find anything similar.

I was wondering if it is possible to rebalance and redistribute optimization work based on fast genetics when some agents finish before others in a generation. I have a local network with several agents, which have different processing power, and some of them, the most powerful, finish the passes of the generation faster than others. It so happens that, with this discrepancy of speeds, these faster agents stop and remain halted until the other slower agents finish the passes distributed to them, before the next generation can be calculated using again all agents.

If it were possible to rebalance and redistribute the remaining work from the slower agents to the faster agents, the total time spent to calculate the entire generation would be lower. To do this, the strategy tester would allow stopped agents (those faster agents) to process the remaining passes of the slower agents.

If this is possible, do we have a place to learn how to rebalance work among agents? I am grateful for the help.

Again, forgive me if this is not possible and if this has already been discussed.

Finally, I would like to suggest to the development team that they implement a UPnP function in MetaTester so that the opening and redirection of agent ports is automatic for NAT clients.

Thanks!
 
plzhelp
 
mlcmc: plzhelp

The reason why no one is answering is because we are not able to help you! There is no such ability that we know of.

Personally I have never experience such a difference when I have agents on different machines, but then my code is usually quite efficient.

So I have never seen such a big difference in time between passes on a fast machine and the ones on the slow one. It always seemed balanced to me!

 
Fernando Carreiro:

The reason why no one is answering is because we are not able to help you! There is no such ability that we know of.

Personally I have never experience such a difference when I have agents on different machines, but then my code is usually quite efficient.

So I have never seen such a big difference in time between passes on a fast machine and the ones on the slow one. It always seemed balanced to me!

I see...

Well, that could be a new feature, I guess. @MetaQuotes, please help!

The thing is that I run an optimization on multiple computers and each one of them run at different speeds, and so an upgrade of a generation is halted until the slowest machine finishes... 

New comment