[Cbe-oss-dev] Real-Time Kernel for the Cell

[Arnd Bergmann]

On Wednesday 26 March 2008, Nicolas Morey-Chaisemartin wrote:

> If I understand well this is due to the Sony software on the PS3 that 
> does not allow real-time?
> Then it should be possible to run a real-time kernel to run some test 
> however performances test would not be significant, right?

It would certainly be interesting to see what the worst case latency
is that you observe on the PPE in the PS3, and use that number for
a soft-realtime OS, similar to the latency guarantees that linux with
preempt-rt gives.

Also note that while you can't do hard-realtime on the PPE when running
a non-realtime hypervisor, you may find it interesting to limit your
realtime tasks to the SPE, which has excellent realtime capabilities
as long as no context-switching of SPEs happens.

The hardest part on the PS3 would probably be finding a way to interact
with the *outside*, which in practice involves either USB, ethernet, or
hypervisor calls, all of which are non-realtime to start with.

For a research project, you can run your realtime code on the SPE, and
use the PPE to simulate outside triggers, but in order to be useful,
you'd need to connect some PCI device with real-time I/O capabilities
that you can access from an SPE directly.

	Arnd <><