Hi Christoph, Which commands are you using to measure performance?
Measuring the startup time of roscore on my Cubieboard2 (ARM Cortex-A7 dual core, 1GB RAM with SanDisk 16GB SDHC Class 10 UHS-1 SD card), I get: *Command used: time sh -c '(roscore &); until rostopic list;do echo "."; done' *Results (four times run: first after cold boot, three others after having killed the roscore process manually): (1) real 0m32.947s user 0m26.700s sys 0m3.410s (2) real 0m22.525s user 0m18.260s sys 0m2.390s (3) real 0m22.560s user 0m18.760s sys 0m2.060s (4) real 0m21.827s user 0m17.980s sys 0m2.320s This surely is not the best way to accurately measure performance, but it gives some idea. If you let me know which measuring commands you are using, I'd be happy to share the results of those on my hardware. Thanks, Kristof On Thu, Dec 5, 2013 at 10:58 PM, Christoph Schultz <[email protected]>wrote: > Hi group, > > I recently joined the group, because I downloaded the meta-ros layer and > wanted to share my experience with ROS on an OpenEmbedded Linux. The > initial compilation went through without any serious issue. So good job :-) > After installing all required packages I could finally start roscore on my > Galileo board. Unfortunately there I ran into some serious performance > limitations. Booting roscore on my EeePC (I use for prototyping) requires > only some seconds. On the Galileo board it took me more than a minute, so I > ran into a timeout. This I could finally avoid by adapting the py-script, > but even when roscore started afterwards I could not query any topics or do > anything with my running ROS. > > One bottleneck was easily identified to be the SD-card connected to the > Quark-CPU. I tried to shift all required files to a tmpfs into RAM and > could achieve a significant speed boost. Unfortunately it is still too slow > to be usable. > > I would be interested in the performance level of your hardware you are > using ROS with. Best case would be, that I finally find out that I just > forgot to move some important files to my tmpfs, so I am facing still the > same bottleneck. Worst case would of course be, that the CPU performance is > just not sufficient for this python based environment... > > Some HW stats of my OpenEmbedded System: > > 400 MHz x86 Quark CPU > 256 MB RAM > 8 GB SD card > > Regards > Christoph >
