Back in the day, I did some experiments on replacing the Parameter Server with Redis.  If the Redis-based system was detected, the clients could switch to the Redis protocol, which has much less overhead.

I'm more dubious in switching the Parameter Server to a full-blown database, or, at least, I think you will have something very different by the end of it.  To me, the Parameter Server is just a config server, and making it better means putting more config there, and making things like Dynamic Reconfigure work better.  One thing that would make it a better config server is to make it possible to shuttle ROS messages back and forth between it, while trying to find the happy balance between namespaces and hierarchical config messages.  ROS is currently sparse on config-oriented messages, and they are one mechanism that could lead to less parameter-related errors, such as misspellings.

If clients need an actual database, they should use a database.  There's many different requirements in choosing a database and it seems a non-goal to have the Parameter Server serve all those possible needs.

HTH,
Ken



On Tue, Apr 16, 2013 at 6:17 PM, Ryan Gariepy <rgariepy@clearpathrobotics.com> wrote:
Personally, I would like the API and launchfile syntax to be kept as simple as possible. Adding a static parameter isn't much extra typing over and above defining a constant, but promotes much more reusable prototype code. I've never personally said "There is too many parameters in this node", but have often said "why is <CONSTANT> not a parameter". Ideally, the dynamic reconfiguration could be taken right from the code, no manual building of a .cfg required.

-Ryan


On Tue, Apr 16, 2013 at 9:07 PM, Gonzalo Abella <abellagonzalo@gmail.com> wrote:
Dear community

My name is Gonzalo Abella. I am going to apply for the Google Summer of Code program. For those who does not know about what I am talking about, GSoC is a program where Google awards stipends to students who successfully complete a requested free and open-source software coding project during the summer. Here you can find more info [1]. There is a project which I find very interesting: 

"Develop new Parameter API for C++ ROS client"
Description: ROS has a concept called parameters, which are settings for individual computational “nodes”. Currently the built-in parameter system in ROS statically sets the parameters at launch time and cannot be changed during runtime. There is an additional system called Dynamic Reconfigure which allows ROS nodes to define dynamic parameters, but this system requires that the parameters be defined outside of the normal ROS parameter API and is not integrated into the ROS C++ client API. This task would be to redesign the built-in ROS parameter system for C++ into a functional prototype which combines the normal static parameters with the Dynamic Parameter’s feature set. There are several features that need to be tackled for this project, for example: having callbacks for when parameters change, lock on access for the parameters (thread safety), grouping of parameters, and how to define the valid ranges for parameter values. The goal is a working prototype which can be used for ROS 2.0 work, but which could be back ported into ROS 1.0 if the new system can be implemented alongside the existing one or the changes to the existing system are minimal.

With this description I have a pretty good idea of the goal of the project. I also have an initial idea of how I would do it. But because this is a community and it is you who are going to use the new Parameter Server at the end of the project, I would like to know your opinions in order to propose a better project (and do it if finally I get selected). Do you miss any feature? How would you improve the Parameter Server?

For example, I think it could be interesting to decouple the Parameter Server from the ROS Master. The reason to do so is to increase its functionality. We could have two implementations and use the one that better suit for you. There is a simple by default implementation. Useful when you don't have many parameters. The other implementation could use a distributed key/value DB. It is useful when you have many parameters and to use the Parameter Server as a database. If the underlying DB is distributed, the DB would be highly scalable. What do you think? Does it fit in the project scope?

I would appreciate any opinion/suggestion/advice/commentary. Thank you very much.

Best regards,
      Gonzalo


_______________________________________________
ros-users mailing list
ros-users@code.ros.org
https://code.ros.org/mailman/listinfo/ros-users



_______________________________________________
ros-users mailing list
ros-users@code.ros.org
https://code.ros.org/mailman/listinfo/ros-users