In year 2008, a feature with the name "timer-based scheduling" (also known as "glitch-free") has been introduced into PulseAudio in order to solve the conflicting requirements of low latency for VoIP applications and low amount of CPU time wasted for handling interrupts while playing music. The novel (at that time) idea was to use timer interrupts instead of sound card interrupts in order to overcome the limitation that the ALSA period size cannot be reconfigured dynamically. This idea turned out to hit some corner cases, and workarounds had to be added to PulseAudio. Despite its age, the implementation of the idea is still not 100% correct. This talk explains why it is the case and what can be done to improve the situation.