Split mythfrontend and mythbackend

Built dedicated mythtv master backend server with Gigabyte GA-7VAXP and Athlon XP 2100+ 1.733GHz (overclocked to 143MHz w/ +7.5% overvoltage). 256MB DRAM. 80GB on Primary IDE, 200GB WD (recent rma) on Secondary IDE. No CDROM drive.

May not need to overclock, as cpu utilization is reasonable during some recording and network serving tests. Exhaust air seems pretty hot.

Mythtv frontend is still ASUS P5P800 (P5P800-UAYZ) and Pentium 4 Processor 550, LGA775 pkg, 3.4GHz, 1MB L2-cache, 800MHz FSB.

Recompiled mythtv on ASUS with:
./configure -–enable-opengl-vsync

Did not use --enable-xvmc because the compile would complain it couldn't find XvMCW libs. Tried deleting references to xvmcw in config.mak but build still broke. Couldn't de-install xvmcw because the DVD programs (xine, mplayer) need it.

Since I don't use xvmc hardware acceleration anymore (can't do deinterlacing), no need to compile it.

Playback of recordings is much better after splitting-off BE from FE. No longer hit cpu bottlenecks when two recordings are in progress (and one is hidef), and we're watching a third show off disk. Seeking forward/backward is more snappy.

But, I did run into nasty IOBOUND errors that would sometimes lockup the backend and require a restart of mythbackend. Measures taken: noapic kernel option at boot, increasing the internal write buffer in RingBuffer.cpp and tv_rec.cpp. These are documented in later blog entries.