Resume:
Computer music researchers have been concerned at least since the 1970s with a fundamental problem: how to build systems that can simultaneously reach high levels of computation throughput, get things done at very short latencies, and offer a clear and consistent programming model (and even, perhaps, a decent user interface).
This talk will address the choices and tradeoffs that beset the computer music system designer: how to use multiprocessors efficiently, how the memory model constrains scheduling; how to manage tasks with multiple, different latency requirements; the costs and benefits of making systems run deterministically; and the interface between sporadic event-driven processes and ones running at fixed sample rates.
Miller Puckette worked at IRCAM 1985-1994, and is now Professor of music at UCSD.
http://crca.ucsd.edu/~msp/