All rapyuta.io docs are open source. See something that's wrong or unclear? Submit a pull request.Make a contribution
Native Network is a beta feature.
A native network allows you to communicate between different ROS environments as described in the following scenarios:
Native networks eliminate the need for creating a separate routed network for the local communication. They also significantly decrease the latency as the communication doesn’t involve a cloud bridge for individual ROS environments communication.
In the case of a native network, multiple ROS environments (all ROS nodes sharing the same ROS_MASTER_URI are considered as a single ROS environment) can discover each other (these could be filtered based on the list of topics/services provided) and the communication happens in a peer-to-peer manner. Each ROS environment has its own ROS master and the rapyuta.io platform uses a sub-component based on the open-source FKIE multi-master to achieve it.
If your ROS package uses the tf topic, and if you don’t add the tf_prefix, don’t expose the topic as there is a possibility of mixing the tf’s. For more information on tf topics, see tf_prefix
Compression, QOS, and Service Timeout are not applicable in the case of native networks.
Communication happens in a peer-to-peer manner using the Fast DDS Discovery Server protocol feature that offers a centralized dynamic discovery mechanism, as opposed to the distributed mechanism used in DDS by default.
Native network server provides a Client-Server Architecture that allows nodes to connect with each other using an intermediate server. Each node functions as a discovery client, sharing its info with one or more discovery servers and receiving discovery information from it. This reduces discovery-related network traffic and does not require multicasting capabilities.
By default, all topics/services/actions are whitelisted in a package component.
RMW_IMPLEMENTATIONwith the value set as
rmw_fastrtps_cppbefore launching your application.
Let’s take an example of 3 ROS packages: package_A, package_B, and package_C that are deployed on the cloud. Suppose we want to establish a communication between these 3 ROS packages. You can use a native network that simplifies and serves as the best medium for communication.
The results are as follows
We can establish this communication by using a cloud-routed network. However, using a routed network involves a cloud bridge for each ROS environment that adds latency when a service/action/topics are being called in the same local area network.
When you subscribe to a topic from a different ROS environment, this subscriber information is kept locally and only shared if a topic is whitelisted in a package component.
When you deploy a native network to the cloud, it is considered as a cloud native network. Any compute resources (CPU/memory) consumed by this native network deployment count against your cloud deployment hours quota.
When creating a cloud native network, the Resource limit field defines the memory allocation and computational capability of the native network. These resources are reserved in the platform for effective ROS communication. You can choose the resource limit of a native network based on the following requirements:
You can connect your deployments to more than one cloud native network for redundancy.
When you deploy a native network to a device, it is considered a device native network. This allows for ultra-low latency peer-to-peer connections between ROS nodes in a Local Area Network (LAN) i.e ONLY deployments on devices on the same LAN can bind to it. Native networks can be deployed to devices that fulfill the following requirements:
To use device native networks the user needs to re-onboard the device to install and enable the new services.
Also, ensure that cpuset and memory cgroups are enabled.
For instance use native networks with a Raspberry-Pi consider adding the following in /boot/cmdline.txt.
cgroup_enable=cpuset cgroup_enable=memory cgroup_memory=1
On reboot, devices configured using DHCP may boot up with a new IP address and the network configuration of a deployed native network becomes invalid. This can be avoided by assigning a static IP to the device you intend to deploy a native network to especially in production systems.