I found that 2.2.18 probably rudely drops samples (lets ocassionally one sample
be played several times) on the Gravis Ultrasound output device. I use 2.2.18
and the native kernel drivers. I wrote this program that should produce a clean
sine tone. Instead I hear a sine interspaced with crackling. The crackling
repeats at 100Hz I guess. There runs nothing CPU-consuming. The sound process
takes 5% CPU. It's funny Linux drops sound samples just at 5% of CPU load. I
got AMD K6-2 3D 400Mhz at 400Mhz, 4*100MHz, running stably between 20-35 deg C
of case temperature. I got PCI / AGP board FIC VA-503+.  There are three serial
ports, none of them was transceiving at that time. There is one ISA NE2000 NIC
which was not sending any packets.  There is a ISA Gravis Ultrasound PnP. Two
IDE disks, one of them sleeping, the second spinning but nod reading/writing.
64MB of 100MHz DIMM SDRAM, which runs for years without problems. 1MB L2 cache.
AGP video card ATI. No X server running, no SVGAlib application running.  No
USB peripherals.

The crackling is not dependent on the buffer size you can set up in the C code.
The crackling is dependent on the frequency of the sine. It's clearly audible
(read: annoying) at 10kHz, audible at 1kHz, inaudible at 100Hz. So I think
they are sample dropouts - the card stops playing and repeats one sample until
kernel gets the breath and whips itself up to supply next audio data.

There are no custom changes in the drivers - no buffer tweaking, nothing. The
Gravis plays modules, midules and mp3 with nearly no problems (apart from when
Loreena Mc Kennith sings too high and too loud, I hear something that looks like
MP3 frames bounds, but I can't surely tell who's responsible for this - if the
Fraunhofer Institute or Linux Kernel)

root@ghost:/usr/src/linux# cat /proc/version
Linux version 2.2.18 (root@ghost) (gcc version 2.95.2.1 19991024 (release)) #14
Mon Jan 29 15:14:27 MET 2001

Clock