For quite some time I've experienced quite high CPU load on my KVM server with an AMD Athlon X2 BE-2400 CPU. All KVM instances used 10-15% CPU time more or less constantly. Idle load was around the 1.5-3 range (reaching over 160 with dynticks).

I've tried Virtio drivers, KSM, dynticks, time sources and lots of other tweaks, but performance have remained sluggish. Tonight I finally hit jackpot!

My server doesn't have hardware constant_tsc or HPET time source only acpi_pm:

cat /sys/devices/system/clocksource/clocksource0/available_clocksource 
acpi_pm
cat /sys/devices/system/clocksource/clocksource0/current_clocksource
acpi_pm

The simple fix is to add the -no-hpet option to KVM and now everything seems back to normal, with each guest only using 2-4% CPU time when idling (using dynticks on both host and guests).

Note 1: Apparently Windows need HPET timing.
Note 2: I had to disable dynticks for the Linux guests again, otherwise the host load is going crazy during the nightly backup job with loads of 30+.
Note 3: Actually setting the kernel parameter clocksource=jiffies is enough to get back old performance without having to disable HPET emulation.

Today I upgraded my three old Linksys WRT54G 2.2 WIFI APs to the latest OpenWRT release 10.03 codenamed Backfire, previously I was running Kamikaze 8.09. The upgrades went almost perfectly smooth.

First I tried the 2.6 kernel using the b43 driver apparently as I had read a successful report. However AP mode didn't work here.

Back on the latest 2.4.37.9 kernel I tried setting up multiple SSIDs on the same device in order to finally set up a guest access network. However it still doesn't work properly and just creates a non working interface.

wl0.1     IEEE 802.11-DS  ESSID:"guestaccess"  
          Mode:Master  Channel:1  Access Point: Not-Associated   
          Bit Rate:54 Mb/s   
          RTS thr:off   Fragment thr:off

It seems like I have an old revision that doesn't support multiple SSIDs, for that you need corerev 9:

# dmesg | grep wl0
wl0: wlc_attach: chiprev 1 coreunit 0 corerev 7 cccap 0x4876a maccap 0x0 band 2.4G, phy_type 2 phy_rev 2 ana_rev 2
wl0: Broadcom BCM4320 802.11 Wireless Controller 4.150.10.5

So while the upgrade to Backfire, didn't really backfire it didn't get me much further either. In the future I might be looking into getting new WIFI hardware that will run OpenWRT an multiple SSIDs. Let me know if you have any recommenations that fits the above.

I'm using Digikam for image management on Linux, using the five-star system to rate my images.

In order to make it easier to move my images between different Linux computers I force Digikam to write tags and ratings to the file's EXIF part. It writes the EXIF tags Windows Rating and Windows Rating Percent.This works out fine for viewing the images on different Linux computers.

Though sometimes I have the need to view them on a Windows computer, and while being an avid KDE user I'm not too happy to install Digikam on Windows.

So I tried to find a decent preferably open source image viewer for Windows that supported the needed EXIF tags for filtering. I've tried Picasa, Irfanview and Xnview so far without any luck:( So I'd be happy for any tips:)

Today I tried upgrading from gentoo-sources-2.6.32-r7 to -r8 on my MythTV server and within minutes I got the usual dmesg output:

DiB0070 I2C write failed

AFAIR it's from my TerraTec T5 USB receiver (I also have a Hauppauge Nova-T PCI card in the server). So I'm back on -r7 for now and checking dmesg with a Nagios plugin. It's so annoying to discover the error when you want to watch a recorded program.

Last week I used a lot of time trying to trace down my puppet clients dying at some point. I could run Puppet ten times without any problems, but suddenly it would just hang without any error message.

It turned out to by Ruby Bug #2739 and after applying the two patches Puppet no longer crashes.

I poked the ruby team on IRC and bugzilla, I'm just filing this here so others facing the same problem will hopefully not waste as much time.