I/O errors
after 2 to 2.5 years. Once this error came, every couple of weeks,
there was no further communication
from the kernel to the drive, requiring hard reboot. Seemed independent of
hdparm settings. Nothing bad showed up
on the manufacturer fitness test and smartctl tests.
Partitioning: be prepared since the FC4 Anaconda graphical installer unfortunately is annoying if you want to customize the partitions. It forces each new partition to be placed at the first place that fits. So you need to trick it by creating dummy partitions which you then remove later before you commit. Eg I wanted 640MB space at start in case I want to put s2d partition there. Here's table:
Device Boot Start End Blocks Id System Size Mount-point /dev/hda1 * 83 95 104422+ 83 Linux 100MB /boot /dev/hda2 96 987 7164990 83 Linux 7GB / /dev/hda3 988 1369 3068415 83 Linux 3GB /usr/local /dev/hda4 1370 7296 47608627+ 5 Extended /dev/hda5 1370 1496 1020096 82 Linux swap / Solaris 1GB swap /dev/hda6 1497 6595 40957686 83 Linux 40GB /home
NVIDIA drivers: Update: This is fine now. The 8178 nVidia driver installs and works fine on the new 2.6.15-1.1831_FC4 kernel; I have not had an X11-freeze in the couple of months since I updated. (begin outdated info) This is quite subtle, and has changed from FC3, due to GCC4.0 I believe. The latest driver which is free of the notorious mouse-able-to-move-but-everything-else-frozen crash bug (which is accompanied by an NVRM Xid error recorded on syslog) which plagued me with the 7xxx drivers, is 6629. With FC4's kernel and 7676 driver, this crash still happens about twice a day, so I need 6629 driver. This is documented on nvnews, with a patch put up by Kel and Aky of WarpX, which has following instructions:
Here are my X11 setup files: xorg.conf regular, xorg.conf.proj for external projectors with max 1024 x-resolution, and xorg.conf.tv for output to TV.
Notes/bugs on FC4 vs FC3, general problems:
Yum repositories: Get this RPM which points your /etc/yum/repo.d files to FreshRPMS. Heed Stanton Finley's warnings linked above about combining Livna with other repos.
Other useful packages:
yum install djvulibre (for .djvu document viewing, creation) yum install freeglut-devel (for GLUT compilation)compile
xxdiff from source (fc4 rpm doesn't talk to
diff correctly), requires:
yum install qt-devel (to compile xxdiff from source) yum install tmake-1.13-1 (to compile xxdiff)
Install via yum,
compat-libstdc++-33 compat-libstdc++-296 compat-gcc-32 compat-gcc-32-c++ compat-libf2c-32so you have access to pre-4.0 versions of GCC.
Install smartmontools via yum, make sure smartd is started at runlevels 2-5, and enable SMART on the main drive: smartctl -s on /dev/hda
Power consumption and SpeedStep: Install the amazing hack speedstep (see my FC3 guide), put in /etc/acpid/events/, in order to switch from 1.13GHz (speedstep -m) to 733MHz (speedstep -b) at will, or let a daemon decide based on load (speedstep -d). It must be executed as root, and even then sometimes gives a permissions warning. By looking at /proc/acpi/battery/BAT0/state I see the following power drainage rates (fluctuations are within 1W):
Heavy CPU usage (floating point): 36W at 1.13GHz
27W at 733MHz
Idle CPU: 23W at 1.13GHz
22W at 733MHz
Since most of the time you're idle, battery life is not increased much.
That was with my wireless PCMCIA card in, and display dim (changing
to bright adds about 2-2.5W).
Little-known fact: /etc/sysconfig/harddisks isn't read anymore! See 157673. I configure hdparm settings now in /etc/rc.local It seems like good DMA settings on main drive are getting set by default now (I get 38MB/s out of the box due to my 7200RPM, which I strongly recommend).
LAPACK seems to be replaced by ATLAS, not a bad move. Have not tested the ATLAS RPM for speed yet, vs the one I compiled in FC3. Append the following to /etc/syslog.conf so you can easily flick over to syslog output to debug stuff:
# writes everything to Ctrl-Alt-F9 console. *.* /dev/tty9
Note ACPI removes the ability to Fn-F1 into Setup or check battery using the BIOS. The only way into the BIOS Setup is at boot time. Kernel boot param acpi=off reverts to APM but then the clock 5-hr jump effect and Gnome battery monitor is messed up, so I still with ACPI.
Suspend to disk/software suspend: not working (screen blank, have to hard-powerdown)
Not tried: power management or laptop modes in linux/Gnome, hdd spin-down.
Cementing my hostname: This is always fun. I want the permanent hostname fricka.localdomain, so FlexLM needed for Matlab can work correctly. I enforce this over the DHCP-chosen hostname by placing the following in /etc/sysconfig/network:
NETWORKING=yes HOSTNAME=fricka.localdomainI also put this in /etc/dhclient.conf so I have a guaranteed fixed alias on the network:
send host-name "fricka";I set /etc/sysconfig/network-scripts/ifcfg-eth0 to have ONBOOT=no and USERCTL=yes
MATLAB: Reinstall 7.0.1, eg from it's original compressed download directory.
cd /usr/local/matlab/sys/os/glnx86 mv libgcc_s.so.1 libgcc_s.so.1_orig ln -sf /lib/libgcc_s.so.1 ./Check with opengl info at Matlab prompt. Thanks to Mathworks support for this fix.
gnupod:
You'll need to
yum install perl-Unicode-String and install
perl-MP3-Info,
then in gnupod directory
./configure and make install.
Works great, /media/ipod mounts automatically and shows up as
an iPod on Desktop. Files are not written until eject /media/ipod
is done. A couple of times the hfsplus has mounted as read only; you will
then need to use hpmount, failing that, is journaling has somehow been
enabled,
plug it in a Mac and do a diskutil disableJournal volumeName.
OpenBUGS
(Bayesian Inference using Gibbs Sampling):
As root, unzip downloaded file in /usr/local,
replace the file LinBUGS by the following:
#!/bin/sh export LD_ASSUME_KERNEL=2.4.1 /usr/local/OpenBUGS/cbugs "/usr/local/OpenBUGS/" "/tmp/OpenBUGS/" "/brugs.so"Then do:
chmod 755 LinBUGS cbugs mkdir /tmp/OpenBUGS chmod 777 /tmp/OpenBUGS ln -s /usr/local/OpenBUGS/LinBUGS /usr/local/bin/LinBUGSExecute
LinBUGS, currently fails unless root!
Not got codes working anyway...
Desktop settings:
hdb: status error: status=0x00 { }
ide: failed opcode was: unknown
hdb: status error: status=0x00 { }
ide: failed opcode was: unknown
hdb: DMA disabled
hdb: ATAPI reset complete
These messages do not seem to correlate with the time of CDROM death.
I do not use kernel boot option hdb=ide-scsi since I gather this
is deprecated. I haven't tested reverting to this much since then
automount ect doesn't work, and kernel can oops if remove module by hand, etc.
My ide-cd is compiled into kernel. Maybe should try as module? I am exploring hdparm options. I have not seen convincing explanation/fix. Please help me!
Contact me at ahb at math dot dartmouth dot edu