When you discount the cost of creating the messages (which has nothing to do with the transport), I have results with nodelets of over 100k messages/second, of any size.  Given some optimization I think this can be increased significantly.  For small messages you're likely to see similar performance over TCP loopback, but we don't have any inter-node metrics at this point.

Josh

On Tue, May 18, 2010 at 9:49 AM, Radu Bogdan Rusu <rusu@willowgarage.com> wrote:
Charlie,

The nodelets are always going to be faster if you can work in a single process, because you never copy data, you just
pass pointers to it. I did some tests with PointCloud2 and PCL, and I successfully managed to pass over 1000 PointCloud2
messages (300k points, xyzw) per second, versus ~20 over loopback. The tests are somewhere embedded in PCL, under tests,
but they are not refined. I'll see whether I have time to clean them up and put the results on a web page.

Josh also did more tests with TCP, and I believe the results are somewhere on the wiki, but I couldn't find them right now.

Cheers,
Radu.

On 05/18/2010 08:28 AM, Charlie Thomson wrote:
> We're experimenting with ROS, and before diving too deeply into it I was
> wondering if anyone had some performance metrics on the messaging. How
> many messages per second can one reasonably expect on modern hardware,
> how many bytes per second, when would it be a good idea to switch over
> to nodelets and bypass lo entirely, that sort of thing.
>
>
>
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users

--
| Radu Bogdan Rusu | http://rbrusu.com/
_______________________________________________
ros-users mailing list
ros-users@code.ros.org
https://code.ros.org/mailman/listinfo/ros-users