[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [rtl] real time systems
On Thursday 05 April 2001 01:06, Cort Dougan wrote:
> It's the case with all the G4's we've tested. Although not an exhaustive
> list of G4's it is a pretty large sampling. We've tested Sawtooth, Yikes
> and the newer model dual G4's. It is possible to not use the IDE at all
> and get much better timing. I've done with with initrd and NFS root.
Ok, but for the applications I have in mind, I'll need to push the drive(s)
about as hard as they can take, but OTOH...
> Do you need better timing that 16us worst-case? If so, I'd suggest testing
> with an alternate card.
>
> If you need better timing than 16us on a workstation a fast x86 should work
> for you.
...I (probably) won't even need to use RTL. The lowlatency patch should do,
at least for UP, and 16 µs should be rather acceptable either way.
(SMP complicates things a bit; it's not acceptable to just multiply the
buffering with the CPU count, because it's "high" already, and it's not
always possible to run the CPUs independently. That is, the per-CPU threads
have to run more accurately in sync.)
That is, the 16 µs shouldn't be a problem for my applications. The excessive
heat (= loud fan noise) and poor efficiency of x86 when executing DSP style
code *is* a problem though; that's why I'm looking for PPC solutions.
(If I needed CPU power only, I could build a cluster of P-III machines, but
the throughput latency requirements makes that impossible to do without RTL
and RT networking. Still interesting, though... Or how about a cluster of
Macs? :-)
//David
.- M A I A -------------------------------------------------.
| Multimedia Application Integration Architecture |
| A Free/Open Source Plugin API for Professional Multimedia |
`----------------------> http://www.linuxaudiodev.com/maia -'
.- David Olofson -------------------------------------------.
| Audio Hacker - Open Source Advocate - Singer - Songwriter |
`--------------------------------------> david@linuxdj.com -'