| ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange |
At the fundamental level, can it be said that ROS is an efficient message passing system, since essentially, that is what it is doing and managing for the most part.. If that is the case, is it possible to use ROS for creating a powerful computing cluster instead of using it only for robotics?
There are other, non-robotics-focused, ways of doing clusters; if all you need is message passing, you might want protocol buffers; if you want to think in terms of computational units, not messages, you want hadoop.
So yes, it will work, but you should start with high-level thinking about the problem you're solving.
I am merely looking for a generic cluster that allows me to model problems as computational units that can run in parallel. They may or may not have huge data sets. Do you think hadoop would help in this case? I always thought it was meant for processing huge data sets in parallel.
I would say this is definitively possible to parallelize algorithms using ROS. In particular, the fact each node runs into a separate process allows to run several times the same node to realize the same treatment on different piece of data.
On the opposite, depending on your needs, your cluster architecture, etc. there is open source alternatives that may be more efficient and easier to use. You may also hit bottleneck that affects nobody else or reveal new bugs, etc.
In short: yes you can. Depending on what you want to achieve at the end, it may be a good idea or a very bad one :)
Asked: 2012-02-27 01:11:10 -0500
Seen: 1,660 times
Last updated: Feb 27 '12
ROS Answers is licensed under Creative Commons Attribution 3.0 Content on this site is licensed under a Creative Commons Attribution Share Alike 3.0 license.
