Hi!

Does there exist a kernel road map for the carambola board?

The reason why I'm asking is because it would be great to get a kernel that is suitable for real time kernel preemption that provides very low latencies / high determinism. This is essential for many kind of machine control and data acquisition applications. The current 3.3.8 kernel is not supported by the real time kernel preemption - some 3.2.x and 3.4.x are supported but not 3.3.x. Please check http://www.kernel.org/pub/linux/kernel/projects/rt/ for available patches.

I have attached a simple jitter test application. It runs for five seconds and logs the jitter. With the 3.3.8 kernel, when the board is idle, then the 95% of the time, the jitter is below 30µs and the worst case is around 80µs. When the carambola is under some load (i.e. ping-flood: ping -f -s 1000 <ipaddr>), then 95% of the time, the jitter is below 140µs and the worst case is around 600µs. A jitter below 1ms is good enough for certain real time application, but not good for what I'm planning to do

regards

Bernhard

Thank you JaXXoN for interesting topic! I am interested in more real-time software as well. If you have any more specific questions, please ask! I will try to answer them or try to get help from our Linux guru. Or you can it by yourself by asking it (submitting issue) on .

We are catching up with latest kernel. Update period is 2 month.

Hi!

Thanks for the quick response!

Please note that the real-time pre-emption patches are typically slightly lagging behind the most recent vanilla kernels (because of the testing, etc.). The latest patch is for 3.6.11. Generally, it appears that the rt patch is only available for kernels with long term support (such as 3.0, 3.2, 3.4. There is no LTS decision for 3.6, yet, AFAIK). So it might be a little bit trick to find a good match. The alternative would be to port the rt patch that comes most close to the carambola kernel in development (i.e. 3.6.11-rt for 3.7). However, that might take some time. We should probably check back with the rt maintainers.

A couple of years back I did some experiments with real time kernel pre-emption with a 400MHz PPC: worst case interrupt latencies had been 55µs (average: 7µs) and worst case user context switch latencies was 84µs (average: 28µs) under heavy load. This PPC part is compatible to the 320 MHz MIPS, so similar number could be expected.

regards

Bernhard