Embedded Linux Benchmarks for Robotics ?
Embedded Linux Benchmarks for Robotics ?
by i-Bot
Tue Dec 13, 2011 12:29 pm
Do you think we might develop a set of Robosavvy benchmarks for embedded Linux boards ?
With so many new boards appearing it is difficult to compare them by specification( MHz, Mb) alone. The work from Marco also suggests that specification alone does not tell a complete story when it comes to video and USB in a robot context with lots of I/O.
I am thinking of a standard configuration.
Linux board with USB port. Plus:
USB Hub
USB2Dynamixel with single attached AX12
Webcam ( Logitech ?)
WiFi dongle (802.11g/n)
USB Memory stick
If the embedded board has in built any of the listed functions, then those can be used instead.
The make and type of hub is likely to effect results too. We might also make comparison there before selecting a reference
Would be good to compare Fit-PC and gumstix to new contenders such as Raspberry Pi and Beaglebone.
What do you think ? Any suggestions on reference setup or tests ?
With so many new boards appearing it is difficult to compare them by specification( MHz, Mb) alone. The work from Marco also suggests that specification alone does not tell a complete story when it comes to video and USB in a robot context with lots of I/O.
I am thinking of a standard configuration.
Linux board with USB port. Plus:
USB Hub
USB2Dynamixel with single attached AX12
Webcam ( Logitech ?)
WiFi dongle (802.11g/n)
USB Memory stick
If the embedded board has in built any of the listed functions, then those can be used instead.
The make and type of hub is likely to effect results too. We might also make comparison there before selecting a reference
Would be good to compare Fit-PC and gumstix to new contenders such as Raspberry Pi and Beaglebone.
What do you think ? Any suggestions on reference setup or tests ?
by PedroR
Tue Dec 13, 2011 2:19 pm
Hi i-Bot
From the work Marco has developed we've learned that the idea "generalized" platform would need to include: Webcam, USB to Serial, WiFi and a fourth port for adding a sensorial interface.
Different boards include different features (Omnima includes WiFi, some OMAP include WiFi and Bluetooth and the number of native serial ports vary as well)
With regards to benchmarking it may make sense benchmarking for specific applications: OpenCV, closed loop controls (ie bus efficiency and latency), etc... [ please suggest more]
Due to different hardware architectures it's really hard to set a "one size fits all" even for benchmark purposes: ebox runs x86 so it's easy to use but lacks behind on FPU (impacting OpenCV); Gumstix for example have more horsepower but add a lot in complexity in terms of cross compiling and available software and so on...
For personal reasons Marco is not with RoboSavvy anymore so it may take a bit to get these benchmarks.
He left us a great foundation and great work (some of it has not been published yet).
This is to say that I'll let Limor know about your ideas and add it to the "to do" list of the person who will eventually replace Marco.
Also let us know what you ha din mind in terms of benchmarking with specific applications or tools.
I trust a BoguMIPS or other simplistic approach will not do it for sure
Regards
Pedro
From the work Marco has developed we've learned that the idea "generalized" platform would need to include: Webcam, USB to Serial, WiFi and a fourth port for adding a sensorial interface.
Different boards include different features (Omnima includes WiFi, some OMAP include WiFi and Bluetooth and the number of native serial ports vary as well)
With regards to benchmarking it may make sense benchmarking for specific applications: OpenCV, closed loop controls (ie bus efficiency and latency), etc... [ please suggest more]
Due to different hardware architectures it's really hard to set a "one size fits all" even for benchmark purposes: ebox runs x86 so it's easy to use but lacks behind on FPU (impacting OpenCV); Gumstix for example have more horsepower but add a lot in complexity in terms of cross compiling and available software and so on...
For personal reasons Marco is not with RoboSavvy anymore so it may take a bit to get these benchmarks.
He left us a great foundation and great work (some of it has not been published yet).
This is to say that I'll let Limor know about your ideas and add it to the "to do" list of the person who will eventually replace Marco.
Also let us know what you ha din mind in terms of benchmarking with specific applications or tools.
I trust a BoguMIPS or other simplistic approach will not do it for sure
Regards
Pedro