On Fri, Jan 19, 2018 at 11:06 AM, benravin <ben.alex@outlook.com> wrote:

just to expand on that a bit. you can build a push model API on top of a
pull model API fairly easily; the other way around is hard.

On a push model on top of pull, pull will overwrite if push is slow, how to
determine how much buffering is required. Any guideline.

You can't know ahead of time. You need to know the behaviour of your non-realtime-safe data sources (e.g. disks).

For reference, in the case of Ardour, we buffer 5 seconds of audio per track by default. This is enough in most cases, but not all; in some cases, it is way more than necessary.