ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
0

Generating communication delay and timing it with ROS subscriber and publisher

asked 2023-03-19 02:04:18 -0500

hunterlineage1 gravatar image

updated 2023-03-21 12:22:14 -0500

I am subscribing to the /odom topic of my turtlebot3 in ROS melodic and Gazebo simulation. I want to do testing on the communication delay with ROS with a constant delay of 50 ms with subscriber and publisher. What are some ways I could do this in my C++ ROS node? Say I want to first subscribe to the /odom topic, then publish a Twist message with a delay of 50ms in addition to the computation time, how would I achieve this? I also want to time the time it takes between first receiving the subscribed /odom topic message, then doing some computation, and then to the time where it gets publish to the cmd_vel topic in the form of a Twist message. What code would I use to time this reliably in my C++ node?

Edit after suggestion in comments: I want to benchmark ROS network performance (ie: inside nodes), or performance of computation (ie: between nodes) via nodes that I created.

edit retag flag offensive close merge delete

Comments

1

Quick note: I would suggest to look at using the tc command (should be part of iproute) instead of doing this at the application level.

gvdhoorn gravatar image gvdhoorn  ( 2023-03-19 04:12:46 -0500 )edit

@gvdhoorn where can I find out more about the tc command from iproute? I tried searching for iproute with ROS and this was the only thing I could find: https://index.ros.org/d/iproute2/

hunterlineage1 gravatar image hunterlineage1  ( 2023-03-19 16:16:02 -0500 )edit
1

It's not a ROS package. iproute is an Ubuntu package. Information about tc can be found in many places, for instance the man page of tc. But there are also many blogs/forum posts detailing how to use tc to introduce artificial latency, package loss, etc to network connections.

I would suggest to clarify whether you want to benchmark your own code, whether you are interested in generic benchmarking of a whole ROS application and whether you want to benchmark ROS network performance (ie: inside nodes), or performance of computation (ie: between nodes).

gvdhoorn gravatar image gvdhoorn  ( 2023-03-20 03:03:31 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
1

answered 2023-03-20 03:52:10 -0500

ljaniec gravatar image

I think this package called message_filters can achieve this behavior quite easily.

The TimeSequencer filter guarantees that messages will be called in temporal order according to their header's timestamp. The TimeSequencer is constructed with a specific delay which specifies how long to queue up messages before passing them through. A callback for a message is never invoked until the messages' time stamp is out of date by at least delay. However, for all messages which are out of date by at least the delay, their callback are invoked and guaranteed to be in temporal order. If a message arrives from a time prior to a message which has already had its callback invoked, it is thrown away.

(...)

The C++ version takes both a delay an an update rate. The update rate determines how often the sequencer will check its queue for messages that are ready to be pass through. The last argument is the number of messages to queue up before beginning to throw some away.

message_filters::Subscriber<std_msgs::String> sub(nh, "my_topic", 1);
message_filters::TimeSequencer<std_msgs::String> seq(sub, ros::Duration(0.1), ros::Duration(0.01), 10);
seq.registerCallback(myCallback);

I highlighted the important bit about possible throwaway messages.

edit flag offensive delete link more

Comments

1

Could you elaborate on how the OP would make this deterministic enough for his "[..] testing on the communication delay with ROS"? The only way I could see message_filters working is by either modifying the sources of all nodes that should undergo testing, or write a dedicated, generic and stand-alone node which uses message_filters to introduce the delay, then republish.

Both options seem like either too invasive, or adding an undesirable non-deterministic aspect to any communication path through the node-graph (ie: it's very hard to schedule async communication and node execution deterministically).

gvdhoorn gravatar image gvdhoorn  ( 2023-03-20 04:06:05 -0500 )edit
1

I assumed modifications to the source code, yes, it is intrusive. The question seemed more basic to me, perhaps only for testing the internal network's ability to carry messages in a personal project. This approach does not survive as general repeatable testing for many different nodes. I would like the OP to comment on what technical level is desired here.

ljaniec gravatar image ljaniec  ( 2023-03-20 05:26:05 -0500 )edit

@ljaniec@gvdhoorn It does seem hard to schedule async communication and node execution deterministically. I am able to make modifications to my source code; it can be as intrusive as needed. The technical level desired here is that I should be able to test the internal network's ability to carry messages with different delay times.

hunterlineage1 gravatar image hunterlineage1  ( 2023-03-21 12:26:50 -0500 )edit

Question Tools

2 followers

Stats

Asked: 2023-03-19 02:04:18 -0500

Seen: 307 times

Last updated: Mar 21 '23