Difference between revisions of "Realtime kernel info"

Latest revision as of 16:23, 12 March 2018

In order to get good performance with audio applications on linux, the system/kernel should be tuned for better realtime performance. This is especially true when using ALSA (or JACK with ALSA) sound devices. The Audioscience sound cards have high quality hardware design and large buffers that make realtime tunning less critical.

There are several ways to get better performance from a linux kernel. They include:

• rtlimits-enabled pam
• realtime-lsm (older approach)

These are links to various resources with more information on realtime performance tunning with linux:

• linux low latency audio information
• article on Ingo Molnar's real time kernel patch
• Linux Realtime Wiki

When do you need Realtime Kernel patches applied?

The general consensus is that a default OS can keep up with the data streaming demands of Rivenedell, however it depends on a number of factors, the primary ones being the sound architecture in use and (sometimes) how critical low latency performance is. If you're running an ASI card, then standard default kernels should be fine -- the cards have huge buffers, and the internal audio routing is done in the card's DSP core (which is hard realtime by definition).

Where it can get dicier is if you're running with ALSA or JACK. If all you're doing is playing and recording, then standard kernels with default settings should work fine. If however you're running a 'virtual switcher' by means of a Local Audio device (where i/o latency becomes important), you may well end up reducing the buffer sizes (either in JACK or in the [ALSA] section in rd.conf(5); in such cases, a realtime kernel may well be necessary in order to avoid xruns.