Hi..... On Fri, Apr 2, 2010 at 04:26, Lourival Vieira Neto <lourival.neto@xxxxxxxxx> wrote: > Hi folks, > > I'm a MSc student, from PUC-Rio, and I've a free software project, > called Lunatik [1]. Lunatik is a infrastructure to extend dynamically the Linux > kernel with scripts written in Lua language. What a nice project you have.....thumbs up! :) Anyway. you may find this URL interesting for dynamic scheduling reading http://kerneltrap.org/taxonomy/term/607 > I intend to create the support to allow users to load schedulers > written in Lua. I think that feature will improve the flexibility and > adaptability of the OS. In that way, it would be possible to load, > dynamically, different algorithms to adapt the OS's scheduler to > different purposes. Moreover, it might be useful to prototyping of new > algorithms. > > What do you think? Here's my not so smart opinion: What you are going to do, is actually something that has been tried several times on different projects (once I remember one, other than plugsched, but I can't recall that for now). But I think the real challenge that you have think first is: how can I make the framework does the scheduling decision fast enough? If it is not fast, than scheduling itself will be the one of the source of the latency. In cpufreq, IMHO this is not so critical because you don't neccessarily need to check the current cpu load and adjust the C and P state everytime. On the contrary, assuming you're using 1000 HZ, the scheduler could be forced to check the time slice expiration every 1 ms and act accordingly. Plus, please CMIIW, Lua is an scripting language, which means it must be intepreted first before it can be executed, thus theoritically it is slower than pure binary code. All in all, I suggest you do some testing first, but see the outcome. At least you try.... -- regards, Mulyadi Santosa Freelance Linux trainer and consultant blog: the-hydra.blogspot.com training: mulyaditraining.blogspot.com -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
