On 8/4/07, Damon Hart-Davis <[EMAIL PROTECTED]> wrote: > Hi, > > I sort of managed to build a custom kernel before, though was a > little short of space on my (SD card) boot/usr area, and have never > applied a whole patch set. > > What I'm mainly concerned about is: > > 1) What's the risk of actually *damaging* something trying to force > HPET and C3/C4? > > 2) What extra power savings might a reasonably expect as a > percentage? If, say, it is likely to be over 25% (eg 5W+) then it > should be worth doing. > > It's all about potential risk/reward! > > (Oh, and I seem to have to repair some of *my* code in the NTP > daemon, so, yes, I probably will try rebuilding it so I can provide > stratum-1 service again, though it's a *long* way down the list of > waker-upers... B^> ) > > Rgds > > Damon > > On 4 Aug 2007, at 16:24, Russell Harmon wrote: > > > My suggestion here would be to make your own kernel. There are lots of > > patches that you can use (hpet, C3) to save power. Also, according to > > http://www.linuxpowertop.org/known.php#ntp NTP generates wakeups when > > it shouldn't. You should compile that from patched source and install > > yourself. Maybe have a look at http://www.linuxpowertop.org/known.php > > and see if there is anything else you can do. > > > > Personally, I find that gentoo can do almost all of that for me, so > > that is what i'd recommend for ease of use (it's what I use). If you > > already have a production system however, it's probably not worthwhile > > to redo it. > > > > On 8/4/07, Damon Hart-Davis <[EMAIL PROTECTED]> wrote: > >> [Sorry if this gets through twice: I messed up last time...] > >> > >> Hi, > >> > >> I have moved a whole bunch of servers off a rack of Solaris machines > >> and onto a single Linux laptop which I am trying to run on low-enough > >> power to run off solar PV for as much of the time as possible (see > >> http://www.earth.org.uk/low-power-laptop.html) and total power use > >> has come down from ~670W to <30W on mains (<20W powered from solar PV > >> via an efficient DC/DC converter). > >> > >> But I'm not sure how far further I can push the h/w I have. > >> > >> For example, it doesn't show C3 or C4 states available which is a > >> shame since it is 99% idle (most time is spent in C2). > >> > >> Also, it doesn't show HPET as being available, which results in ~50 > >> extra wakeups per second. > >> > >> Now I *am* running a bunch of DNS, NTP, SMTP, and HTTP servers on > >> that (Core Duo T2250) machine, albeit quiet ones, that generate ~30+ > >> wakeups per second themselves legitimately. > >> > >> Is it worth trying to force activate HPET, C3 ad C4? > >> > >> How can I safely do it if so? I have looked at the BIOS with > >> dmidecode and biosdecode, and don't get any great clues. > >> > >> I have played with building a previous kernel but am now simply > >> running a stock (22-rc7) Ubuntu "Gutsy" kernel in my "Feisty" system. > >> > >> I am using cpufreqd/ondemand which I have tuned as good as I can get > >> (quick builds, idles in lowest freq ~99%). > >> > >> I am using laptop-mode. > >> > >> And of course I have used powertop to help tune userspace and > >> cpufreqd, which is why I am here! > >> > >> Rgds > >> > >> Damon > >> > >> > >> _______________________________________________ > >> Power mailing list > >> [email protected] > >> http://www.bughost.org/mailman/listinfo/power > >> > > > > > > -- > > Go read my blog: http://eatnumber1.blogspot.com/ > > > > I am very far from an expert on C3 or HPET, but I don't believe that there is any risk of damaging your hardware by applying the patches designed to force C3. I'd suggest researching if said patches are designed for your hardware first however.
Also, i'm not sure about percentages, but I understand HPET alone can provide a significant reduction in wakeups. ~Russ _______________________________________________ Power mailing list [email protected] http://www.bughost.org/mailman/listinfo/power
