[RDD] Real time Kernel

Fred Gleason fredg at blackhawke.net
Tue Aug 12 17:44:43 EDT 2008


On Tuesday 12 August 2008 04:28:36 pm Dan Mills wrote:
> I am however not convinced that anyone except some lunatics pushing
> latency the limits with jackd really benefit from an RT kernel as
> reasonable sort of RD latencies (1024 frames, two or three periods)
> should be vastly greater then the sorts of kernel latencies even a non
> RT kernel should be experiencing).

For most current systems I've seen, I absolutely agree.  For example, I can 
run RD with the LiveWire driver (eight in, eight out) even on rather modest 
P4 hardware with stock SuSE kernels with no xruns whatever.  The big resource 
needed is RAM -- that much I/O does chew up memory for ringbuffers!

There still are some occasional oddball systems out there where latency can 
become a problem though.  Often the cause seems to be misbehaving hardware -- 
video cards with overly aggressive bus-mastering schemes used to be notorious 
culprits.

Cheers!


|-------------------------------------------------------------------------|
| Frederick F. Gleason, Jr. |               Chief Developer               |
|                           |               Paravel Systems               |
|-------------------------------------------------------------------------|
|    Optimization is fifteen percent science and eighty-five percent      |
|    black magic.                                                         |
|                                         -- Linus Torvalds               |
|-------------------------------------------------------------------------|


More information about the Rivendell-dev mailing list