Simple Network Simulator (SNS)

The Simple Network Simulator (SNS) aims to provide simple and fast capabilities for the transmission of V2X-messages using Ad-hoc communication. In order to stay performant the simulator makes abstractions in certain places. Those abstractions will be discussed later on.


This simulator does not need to be installed. It is delivered as part of the Eclipse MOSAIC installation package.


The SNS offers some configurability regarding the way transmissions are simulated.

Main Configuration:

Parameter Description type Default Value
maximumTtl Defines the upper bound for the amount of hops a message can make. (Note: messages can have individual ttl’s) int 10
singlehopRadius Fallback radius to be used for transmission, if no radius is defined in the AdhocConfiguration double 509.4
singlehopDelay A delay configuration for the direct communication between two nodes. ( See here) Delay ConstantDelay
singlehopTransmission This contains the transmission configurations for lossProbability and maxRetries. CTransmission n/a
adhocTransmissionModel A class extending AdhocTransmissionModel, this will decide the logic for transmissions. AdhocTransmissionModel SimpleAdhoc TransmissionModel

On default the SNS will use the SimpleAdhocTransmissionModel with a ConstantDelay using 0 as delay. This means it usually makes sense to specify the AdhocTransmissionModel explicitly and use a more realistic Delay.

Example Configuration:

    "maximumTtl": 20,
    "singlehopRadius": 300.5,
    "singlehopDelay": {
        "type": "SimpleRandomDelay",
        "steps": 5,
        "minDelay": "1.5 ms",
        "maxDelay": "2.5 ms"
    "singlehopTransmission": {
        "lossProbability": 0.0,
        "maxRetries": 0
    "adhocTransmissionModel": {
        "type": "SimpleAdhocTransmissionModel",
        "simpleMultihopDelay": {
            "type": "GammaRandomDelay",
            "minDelay": "10 ms",
            "expDelay": "30 ms"
        "simpleMultihopTransmission": {
            "lossProbability": 0.1,
            "maxRetries": 2 

Read the detailed documentation of the SNS Configuration.

Transmission Logic

SNS differentiates between two types of Ad-hoc transmissions, geographically- and topologically-scoped transmissions, which generally are abbreviated with GeoCast and TopoCast respectively.
GeoCasts are limited to BroadCasts. Accordingly, there is no explicit addressing of receivers (other than, instead a destination area is specified. However, GeoCasts allow for multihop forwarding.
TopoCasts on the other hand use means of IPv4 addressing to transmit messages. Since the SNS was not build to simulate transmissions using complex topology-constructs, TopoCasts are limited to transmissions with a single hop. However, TopoCasts support BroadCasts and UniCasts (we are omitting Anycasts). Most transmissions in the Ad-hoc domain will be some form of Broadcast, meaning every reachable entity is eligible to receive a message.

Transmission Type
or Unicast?
Simulate TopoCast using
singlehop configuration.
Simple transmission
Simulate transmission using
Sender directly able to reach
destination area?
Flooding in destination area
Greedy Forwarding to reach area,
then flooding in destination areay.
This flowchart tells how different types of messages are handled internally.


The only way of directly addressing entities is a SingleHopUniCast (see figure below), the sender will try to address an entity in its transmission range.

SingleHopUniCast: The RSU is directly addressing the green vehicle.
The counterpart to that is a SingleHopBroadCast (see figure below), this form of transmission is commonly used for CAMs (Cooperative Awareness Messages) and other types of intermediate warning messages to all entities in transmission range.
SingleHopBroadCast: The RSU is addressing all units in transmission range.


As already explained, GeoCasts do not support direct addressing, so there is no form of UniCast. Instead of addressing entities, GeoCasts specify a destination area in which a message should be distributed. The SNS supports two ways to simulate GeoCasts. A simple but performant model (SimpleAdhocTransmissionModel) & a fairly realistic model ( SophisticatedAdhocTransmissionModel).

The simple model assumes a transmission to all entities in the specified area, whereas the delay will be calculated using the configured delay-type and the successful reception will be determined by the uniformly distributed lossProbability. The figure below depicts this behaviour

Simple GeoBroadCast: The RSU is sending to all entities in the destination area. All arrows (transmissions) will have a uniquely calculated delay or possible loss.

The realistic model accounts for possible transmission failures more accurately. The easiest case is that the sender itself is inside of the destination area (or is able to communicate with an entity inside the destination area) and will start a Flooding Transmission within this area (see figure below).

GeoBroadCast using Flooding Transmission. Note: the area is not limited to circles.

In case the sending entity is outside of the destination area, a Forwarding Transmission has to be executed first. This is can also be described as an AnyCast, since the goal of this transmission is to reach any entity inside the destination area. We try to achieve this by building a “chain” of entities, that will forward the message to the destination are (see figure below).

Forwarding Transmission, by building a “chain” of vehicles.

The SNS however never uses Forwarding Transmissions individually, rather they are combined with a Flooding Transmission, which will simulate a way, that GeaCasts can be implemented in reality. The figure below depicts this behaviour.

Forwarding Transmission followed by a Flooding Transmission to realistically simulate GeoCasts.

Transmission Models

As already mentioned in the previous abstracts, the SNS supports different transmission models for different use cases. Depending on the configuration of the SNS and the type of message send, different models will be used. The models are located in the package org.eclipse.mosaic.fed.sns.ambassador.model. This chapter aims to give a detailed inside in the workings of the models.


This is the most basic of all transmission models and will be your model of choice if you are not interested in completely accurate transmission results but care for performance. This model will approximate GeoCasts using the defined simpleMultihopDelay and simpleMultihopTransmission parameters. For TopoCasts the usual singlehopDelay will be used. This model only checks, whether a potential receiver is inside the destination area and has enabled Adhoc capabilities. If those conditions are met it will simulate the transmission by calculating an actual delay value and saving it into a transmission-result. Such a result holds information of the success of the transmission, the delay-value, the amount of hops, and the number of attempts. Though the amount of hops will always be 1 for this model.


This model offers are more realistic simulation of adhoc transmissions, using an implementation of a greedy-forwarding and flooding algorithm (see greedy forwarding & flooding). For TopoCasts this model behaves very similarly to the SimpleAdhocTransmissionModel, since TopoCasts are always configured with only one hop. For GeoCasts however, this model follows the flowchart above, trying to “approach” a destination area if it can’t be reached directly.

Approaching (Greedy forwarding)

Approaching can be imagined as building a “chain” of entities to reach an area. However, there is no guarantee, that even if such a chain exists, it will be found. The way that this chain is build follows the subsequent steps:

  • Start from the sender and collect all reachable entities.
  • Choose out of all reachable entities the one, that is closest to any node in the destination area.
  • Use the chosen node and repeat the first step.
  • Repeat until either a node inside the destination area is reached, or the TTL (time to live) is exceeded.

By always choosing the node with the shortest distance to the destination area, we omit a lot of possible solutions. Greedy Forwarding isn’t optimal, but offers a performant approach for this problem. “Face Routing”-algorithms will always find a path if one exists, however this hasn’t been implemented yet (feel free to contribute :). The figure below shows an example of those shortcomings, the message will be send using the green nodes and won’t receive the destination area, even though there is a possible “chain” using the yellow nodes.

image/svg+xml Layer 1 Destination Area Destination Area Destination Area Sender Sender Sender
This figure depicts a case were the Approaching Transmission wouldn’t reach the destination area, even though there is a possible way. (The dashed lines represent the communication range)


The implementation of Flooding is fairly equivalent as described on Wikipedia. Each entity forwards the message to all entities in its communication range. Entities, that already received the message won’t receive it again; this is different from many real-life implementations, where messages are send to all reachable entities except the sender. However, since the simulation has total knowledge of all simulated entities, it is easier to overcome a lot of the disadvantages, that flooding faces in real world implementations.

Implementing your own AdhocTransmissionModel

If the implemented models don’t suffice your needs you can easily implement your own. Create a class extending AdhocTransmissionModel and implement the abstract methods for sending TopoCasts/GeoCasts. A possible extension could be to allow for multihop TopoCasts, building an actual topology and transmit your messages using that topology. Also, the aforementioned “Face-Routing” could be of interest. Additionally, the calculation of delays could be made more realistic.

Accessing SNS-functionality from your applications

To enable SNS in your scenario follow the steps as described in Scenario Configuration.
An overview of how to configure AdHoc-modules and usage of the API for Routing and Message-Building functions, can be found under Application Development.