Date: Fri, 14 Nov 2008 09:55:29 +0900 From: "Cyrille Szymanski" <cnszym@gmail.com> To: "Alexander Motin" <mav@freebsd.org> Cc: FreeBSD acpi <freebsd-acpi@freebsd.org> Subject: Re: RFC: powerd algorithms enhancements Message-ID: <ba5115170811131655i463050c1rd17af354f51f3e6a@mail.gmail.com> In-Reply-To: <49189B47.7090509@FreeBSD.org> References: <491404BF.1090807@FreeBSD.org> <49186111.9020103@root.org> <49189B47.7090509@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
2008/11/11 Alexander Motin <mav@freebsd.org>: > Universality and predictiveness. Systems with 2 and with 32 levels will > drop/rise frequency with same absolute rate. If we are trying to speak about > some comparing, algorithm should be hardware independent. Hi The heuristics PAST and FLAT are two examples of strategies which do just this : they determine the desired operating frequency (in Hz) and then switch to the closest level available. >> There's a patch Cyrille Szymanski has sent me to review that implements >> the FLAT and PAST algorithms in powerd. I think we should not add new >> modes that are heuristics (including this one) until we have a chance to >> compare it to algorithms that have been the result of real research. > > So show it to us. This area is heuristic by default as there is several > opposite criteriums, so any algorithm will be heuristic. Although I see the benefits of your proposal, and reckon that systems with may levels are more and more common these days. The point is, I think, to avoid adding tuning knobs to powerd that would be too specific to your configuration and which might benefit everyone. Research will (I hope) indicate the way to go. Thanks ! -- Cyrille Szymanski
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?ba5115170811131655i463050c1rd17af354f51f3e6a>