Date: Fri, 17 Jun 2005 22:48:20 +0200 From: Richard Mittendorfer <[EMAIL PROTECTED]> Subject: Re: [LIB] Re: Back, with Linux!
-- RESEND -- (the first one didn't came through?) Also sprach Christopher Kalos <[EMAIL PROTECTED]> (Wed, 25 May 2005 14:40:57 -0700): > On May 25, 2005, at 4:10 PM, Richard Mittendorfer wrote: > >> [...] > > this kdrive sounds very interesting. but you need fb to use it? > > maybe i'll test it out. > > You need fb to use it on the Libretto. Their VESA server doesn't seem > to work right, and they didn't build a neomagic version. i did a little google and it schould be possible to build a neomagic driver in. but the driver i found only supports some minimal acceleration[1] (solid and bitblt). i think i'll stay with XFree. [1] http://cvs.freedesktop.org/xserver/xserver/hw/kdrive/neomagic/README?rev=1.1&view=markup update: tried an glibc compiled neomagic tinyx. fast, quite small, but lacks some features. > > [...] > > as i still have the original TOSHIBA MK4310MAT in the lib and i > > usually have 70% full with oggs, so i'm in need of some extra space. > > nfs is IMHO > > the fastest/best solution here. i'm going to buy a bigger drive > > soon, but can't afford it now. > > I see your point, I'm just one of those people who was brought up with > > an innate distrust when it comes to NFS. :) ACK version 4 will implement secure RPC (kerberos V or SPKM3). i access my nfsshares via tunnel, so it's a quite trusted here. > >> > >> The Fedora kernel, which I've rebuilt at times, doesn't seem too > >> cooperative on the Libretto, even though I've dealt with FC1 and > >FC2 > on various systems, all the way up to Dual-proc Opterons. On > >the > other hand, it has EVERYTHING I need available as a module, > >including > Toshiba ACPI support, Cardbus, and so on. > > > > i have never had fedora installed, but you can take a vanilla from > > [1] if you think it's fedora's fault. i know the libretto working > > with 2.2.20, 2.4.18 and .24, 2.6.8-11. i've put the configuration i > > use (2.6.11) to [2]. some more info at [3]. HTH > > > > [1] http://www.kernel.org > > [2] http://www.mittendorfer.com/rm/temp/lib110ct_kconfig.txt > > [3] http://www.mittendorfer.com/rm/temp/l110ct.txt > > > > btw: this use of cflags is quite debatably, but i use -O3 > > -fforce-mem -ftracer -pipe -fexpensive-optimizations when i compile > > 2.6 (and some 2.4) kernels. IMHO they are faster then.. > > i dont't use this flags on productive systems only privatly and > > havn't seen any problems: > > > > [EMAIL PROTECTED]:~$ uptime > > 21:57:41 up 142 days, 1:41, 3 users, load average: 0.82, 0.29, > > 0.28 > > [EMAIL PROTECTED]:~$ > > Also, in the end, the whole thing's barely worthwhile to me. It runs, > it's a hair slower than it could be, and sure, some stuff could go > from modular to built-in for a meager boost, but it's a P-233, not a > P-66. I honestly believe that I have more than enough clock cycles > when I trim the really fat stuff, like a RAM-chugging > overly-dependent GNOME (or KDE) setup. I certainly don't care enough > about the Libretto's stability to worry about aggressive > optimizations, but I would love a faster startup time for X/fluxbox. even if you optimize everything and compile with "perfect fit" it won't give you more than 5%. i also tweaked around in /proc/sys/ and maybe found a few percent there - it's only worth the work to understand what's going on in the kernel. just wanted to say: i'll _never_ be perfectly satisfied with performance ;) maybe prelinking will decrease your startup times. > > [...] > > ide performance is that bad on the lib - a good scheduler helps. > > messured with hdparm -t shows the old 2.4. ide driver still (lill' > > bit) ahead of the 2.6 elevator. > > > > Yeah, same on Opteron, not so much on Xeons, and NPTL on the Libretto > may or may not be a big deal. havn't tried on the lib yet, but my other boxes do well with the new glibc. > >> GKrellm indicates a temp of about 60 degrees Fahrenheit under > >normal > situations, spiking to around 140 if I don't force ACPI > >throttling to > 100CT clock speeds. > > > > cat /proc/acpi/prozessor/CPU0/throttling > > the 110ct seems to have 15 throttling states. what do you use to > > throttle the cpu. AFAIK cpudyn or cpufreq don't work on the moblie > > pentium mmx, do they? > > echo [value] > /proc/acpi/processor/CPU0/throttling i see :) > I go for around 4 when I'm not in need of full power, though I'm > thinking of tossing in an ACPI hook to drop it to 15 or so, disable > any inserted PCMCIA devices, and turn down the sleep timing for > /dev/hda to 5 seconds, for a sleep mode without the lag of sleep > mode. Of course, the ACPI Lid hook sucks so bad that I doubt it's > worth the trouble. it takes 6 seconds to sleep with a cardbus wifi plugged in. exactly the same time to come back out to normal operation. that's fine for me. i don't do any automagic cpu throttling since temperature seems quite ok here. i ususally have sleepd calling a selfrolled script /usr/local/bin/pwr to put the lib to suspend if i forgot to do it manually. this script also does the necessary thing's like switching off the backlight, calling sleep-states or setting spindown times. in combination with the laptop-mode-tools (some /proc/sys/vm/ entries to prevent harddisk access and more) this results in very usefull notebook behavior. the acpid takes care of events like LID and buttons. usually the powerbutton is configured to shut down (init 0) - that was anoying, i turned this off. now LID and powerbutton wake it up. by grep'ing the LID's state i decide what actions to set (to not get too hot if the lid is closed or to suspend if the soundcard irq isn't called). > > [...] > > how to you read the temperatures? if you get them via ACPI, can you > > please post the dmesg part about ACPI and > > > > $ cd /proc/acpi/thermal_zone/THRM > > $ for i in *; do echo -n " $i = "; cat $i; done > > > > thanks a lot - i really want to get the thermal part working. i have > > tried various kernels, some selfmade patches and also tried a custom > > dsdt - no avail. > > > > It's coming from ACPI, alright, and for the moment, it seems stuck at > 140 F. It's using the toshiba_acpi module for 2.6, though, which is > supposedly trustworthy. I'll get back to you once I'm sure I've got > valid output. i should try this toshiba_acpi module. i deselected it because it doesn't seemed to give any additional features. i use "toshset" for all that bios stuff and ac/battery/cpu looks well supported with plain 2.6 acpi. update still no temp avail. i still wonder if there actually is one. > [...] > > i prefer openbox or fluxbox. xfe as filemanager. > > > > I'm sticking with fluxbox, as I'm learning how to use its tabbing, and > > subtly tweaking the environment as needed. The lack of a decent SSL > capable browser and IMAP client are killing me, though. the ~/.fluxbox/keys file is useful. i edited it to open a terminal, sleep, suspend, mailcheck... with "ctrl+alt+<key>". easy. functional. > CK sl ritch
