Huerzo Posted January 7, 2005 Report Share Posted January 7, 2005 Link to story OK. I admit. Know I'm talking about an area I'm not completely familiar with, but I've some experience. I read the article and I came up with some ideas. Whilst Genetic Algoritms may seem just another nice (and funny) tuning technique, I'm having a view that they'll playing a major role when optimizing Linux kernel (and other program code). Why? 1. They will always produce the best possible result (if given time enough) 2. They fit themselves to different environments and still produce the best possible result -> same code, manual tweaking not necessary (OK, now I broke somebody's heart :lol:) So, eventually they are saving much both developers' and users' time. At least I'm fascinated by the vision that computer themselves are doing the dull and dirty work of tweaking softwares to approach better computing. For example, there's been a lot of talk that the boot-up time of linux is too long. And I've read that there will be done some research what's the best possible order launcing processes while booting system up (article). I see this could be an excellent chance to use Genetic Algorithms to find out the right order. The series of processes could be one "chromosome string" and one process one chromosome in that string. First mixing processes to random order, then using UML (User mode linux) to see if "the mix" works. Of course, timer is running to find out how long the boot will take, Then best chromosomes are mixed together to reach the optimal solution. And new evolution rounds until we have an optimal process starting sequence. Of course this is rough and simple view, I've never compiled a kernel (or tweaked it), so I guess much more details are needed to take on account before "something" happens. And much time. But I guess this could be done. Or have I got something wrong? Quote Link to comment Share on other sites More sharing options...
aru Posted January 8, 2005 Report Share Posted January 8, 2005 Looks pretty interesting, it's something to add to the things-to-read pile Quote Link to comment Share on other sites More sharing options...
a13x Posted January 17, 2005 Report Share Posted January 17, 2005 (edited) If you are interested in testing the genetic algorithm get yourself the latest kernel sources, 2.6.11-r1, plus the 2.6.11-r1-nitro0e patch and compile it. IMHO the nitro patchset is the best around. Not only it will make your kernel run faster and better but includes bleeding edge stuff too, a la Gentoo. ;) Here's the link: Nitro I'm testing it right now. Here are my results: Alex ~ # cat /proc/genetic/as-ioscheduler generation_number: 46 num_children: 8 child_number: 6 num_mutations: 8 avg_fitness: 40743 last_gen_avg_fitness: 5142 Fitness history generation(30): 8666 generation(31): 3365 generation(32): 7805 generation(33): 6028 generation(34): 11129 generation(35): 7120 generation(36): 7028 generation(37): 8385 generation(38): 7102 generation(39): 12384 generation(40): 8068 generation(41): 29428 generation(42): 40714 generation(43): 9028 generation(44): 6857 generation(45): 5142 Alex ~ # Edited January 17, 2005 by a13x Quote Link to comment Share on other sites More sharing options...
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.