8 Apr
2005
8 Apr
'05
6:32 p.m.
On Fri, 2005-04-08 at 14:50, Jack O'Quin wrote:
Jan Depner <eviltwin69(a)cableone.net> writes:
Please, don't think I was being disparaging. I am absolutely
thrilled with the kernel work that has been going on. This thing
rocks! The last I heard (from contractors working for us who were doing
soft real-time on - shudder - Windoze) M$ can only get you to about 30
milliseconds.
Jan
On Fri, 2005-04-08 at 12:37, Lee Revell wrote:
> Personally I think you're wasting your time. Ingo's RT preempt patches
> let you do hard realtime with Linux using the existing driver base.
I'm not sure why he's trying to do
this because for audio the
latest patched kernels appear to be more than adequate. That said,
I don't think you can get a guaranteed 15 microsecond interrupt
response with anyone's patches to the standard kernel. I wouldn't
call what we have now hard real-time. More like soft real-time.
Maybe when MontaVista gets done...
Agreed.
The Linux kernel will always be "soft" realtime. They can never
*guarantee* specific interrupt response latencies, because too much
kernel code can come along later and mess it all up by disabling
interrupts or holding locks.
What many people overlook is that for most practical purposes, soft-RT
is what they really want. Hard-RT is only appropriate for a few
specialized imbedded systems.
Thanks to the hard work of Ingo, Lee and many other folks, recent
Linux kernels provide excellent soft-RT, right out of the box. With
Ingo's RT preempt patches, the results are world-class. There will
always be "two steps forward, one step back" regressions, but those
fixes keep getting migrated into the base kernel, and the general
trend is quite encouraging.