Overview

This package includes three separate tests benchmarking the speed of msg passing in various contexts.

  • Passing a copy of float64 between two processes.
  • Passing a copy of float64 between two threads.
  • Passing a pointer to a float64 between two threads.

all of course, via topics. It is also worth perusing this code just to make sure you're clear on the syntax which separates the above passing methods, it can be subtle if you're new to it.

Usage

For all of these, make sure you have pre-run roscore.

InterProcessCopy

./talker
./listener

It will output latencies and finally when terminated (ctrl-c) will output the average latency to standard output.

InterThreadCopy

./thread_passing_copies

InterThreadPointers

./thread_passing_pointers

Note this one is almost functionally identical to the previous one - you just pass a boost shared pointer when publishing instead of an actual object.

Notes

Passing inter-thread with pointers is fast. There is no tcp/ip layer and no copying. Also, you may find on very low spec boards, the latencies drastically increase - there are many polling loops in the ros and this seems to play havoc with the msg passing latencies. Also make sure you have everything optimised (release modes) and the floating point settings correctly set.

Wiki: eros_msg_latency (last edited 2010-10-06 08:04:13 by DanielStonier)