0
$\begingroup$

Rosanswers logo

I'm trying to decide between running ROS Indigo on a single board computer like a Nvidia Jetson TK1, Inforce 6410 Plus or an Odroid vs. using multiple low-cost single board computers like the new Raspberry Pi. The reason I'm looking for a very fast board or multiple boards is that I want to run multiple packages that are computationally expensive like vision, learning, SLAM, etc. I've heard that using distributed processors is pretty easy given the support in ROS so I'm not sure which approach is wiser. While ROS can allow some packages to run on different boards, I'm guessing that only a single board can be assigned to a specific package like vision. So, which approach would be the wisest?


Originally posted by d7x on ROS Answers with karma: 53 on 2015-07-25

Post score: 1

$\endgroup$

1 Answer 1

0
$\begingroup$

Rosanswers logo

One thing to consider is if nodes need significant data transfer between them (point clouds etc..) then having each process running on a different computer will mean lots of traffic on the network - nodes that all need the same data running on different computers will mean duplicating the data over the network. The raspberry pi does not have great usb, and the ethernet adapter uses it as well, so I would be cautious about using lots of raspberry pis networked together. The ethernet on the odroid boards is better (they have a dedicated connection to the processing chip rather than sharing usb).

Having stuff distributed in ROS is relatively easy, but it makes sense to try and divide things up into nodes that share the same data, and run them on one device if the data is large, with higher level messages (less frequent and less data) communicating between devices where possible - its classic parallel processing issues.


Originally posted by nickw with karma: 1504 on 2015-08-02

This answer was ACCEPTED on the original site

Post score: 1

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.