[Tutor] OpenMP

Ahmed AL-Masri ahmedn82 at hotmail.com
Sun Oct 10 06:45:59 CEST 2010


 Thanks for fast responding. I will try to use the threads and see how the
 performance would be.. actually I am using that for my artificial neural
 network and the problem is regarding to the ANN limitation when I used a 
big
 no of inputs. so one way to overcome this problem is by distributing and 
now
 I have like 3 networks in my system with slow processing. May be parallel
could have little effort.


> Depending on how long running these processes are you may be able
> to separate them out completely into separate server processes
> in true client server mode. Effectively creating a network process for
> each network then have a load balances so that each incoming
> request gets sent to one of the server processes. That way you
> can scale linearly by adding moreservers as required (the load
> balancer starting a new process each time the number of active
> requests passes a given limit.) This is how many uindustrial scale
> databases handle high processing loads, each SQL request is validated
> and if OK passed to a query server, the queries are distributed over
> the available servers to ensure even loading and each server can
> run on separate CPUs as needed (but still controlled by the OS).
>
> The downside of this is of course the complexity of writing the
> loadbalancer which must track all incoming requests and which
> server they are on and also handle the responses from the servers
> when complete, making sure they go back to the right source.
> Its a little bit like writing a web server... or at least the listener 
> part.
>
> HTH,
>
> Alan G.
>
> 


More information about the Tutor mailing list