As a proof of concept for the basic solution, I proposed a java algorithm.
However, since the beginning my aim was to implement the solution on a cpu with no OS, I am targeting the ARM Cortex Sitara, but I thought the LM4F was already a good starting point.
The basic idea is to use low cost / low power cpus in some kind of network enabling them to cooperate together.
Would it make sense to consider an MCU, such as the Stellaris?
Indeed its 50MHz frequency does not keep up with i5 /i7 CPUs at 3+GHz, but in some cases it might be enough.
Admittedly my algorithm is not really optimized, but for this experiment it is probably ok, in the end my idea is to verify the possibility of having a network of smart sensors, able to integrate AND PROCESS data , without the help of an external dedicated host.
So far, it's just for fun, curiosity.
Today I simply copy&pasted my java algorithm into CCS, converted it into C and redirected the System.out.println() to UARTprintf(), but most of the code recompiled without intervention.
You can download the full source code here (requires StellarisWare installed)
Obviously, the result is the same, as you can see from the console of the java program running on my PC and the RealTerm window that captures the serial output of the Launchpad.
Did not measure performances yet, however it is absolutely clear that the PC is WAY faster in getting the solution, even if java is not nealry as efficient as ARM C.
Comparing a multi core 2.6GHZ 64 Bit CPU with a single core 50MHz 32 bit one is not exactly what we would call a fair match.
Still, the Launchpad managed to get the solution.
It is indeed a non conventional usage of an MCU, I did not use any peripheral (excluding the UART and the onboard LED, for debug purposes), I simply used it as a CPU.
Since now I have the basic algorithm working on the Stellaris, I can start thinking of allowing multiple parallel CPUs working at the same time.
Stay tuned!
No comments:
Post a Comment