Skip to main content
Version: 1.9.3

Using network

Introduction

GAMA provides features to allow agents to communicate with other agents (and other applications) through network and to exchange messages of various types (from simple number to agents). To this purpose, the network skill should be used on agents intending to use these capabilities.

Notice that in this communication, roles are asymetric: the simulations should contain a server and some clients to communicate. Message exchanges are made between agents through this server. 6 protocols are supported (TCP, UDP, MQTT, HTTP, Websocket, Arduino):

  • when TCP, UDP or Websocket protocols are used: Agents can be either clients or server depending on the needs of the simulation.
  • when the MQTT protocol is used: all the agents are clients and the server is an external software. A free solution (ActiveMQ) can be freely downloaded from: http://activemq.apache.org.
  • when HTTP is used: the agents can interact with webpages/webservices through raw GET, POST, PUT, and DELETE requests
  • when arduino is used: one agent of the simulation can connect to an Arduino as a client.

Which protocol to use ?

In the GAMA network, 6 kinds of protocol can be used. Each of them has a particular purpose.

  • MQTT: this is the default protocol that should be used to make agents of various GAMA instances to communicate through a MQTT server (that should be run as an external application, e.g. ActiveMQ that can be downloaded from: http://activemq.apache.org/),
  • UDP: this protocol should be limited to fast (and unsecured) exchanges of small pieces of data from GAMA to an external application (for example, mouse location from a Processing application to GAMA, c.f. model library),
  • TCP and Websocket: these protocols can be used both to communicate between GAMA agents in a simulation or between GAMA and an external application.
  • HTTP requests: this protocol should be used to communicate with an external webservice.
  • Arduino: this protocol should be used to communicate with an arduino device

Disclaimer

In all the models using any network communication, the server should be launched before the clients. As a consequence, when TCP, Websocket or UDP protocols are used, a model creating a server agent should always be run first. Using MQTT protocol, the external software server should be launched before running any model using it.

Declaring a network species

To create agents able to communicate through a network, their species should have the skill network:

species Networking_Client skills: [network] {
...
}

A list exhaustive of the additional attributes and available actions provided by this skill are described here: network skill preference page.

Creation of a network agent

The network agents are created as any other agents, but (in general) at the creation of the agents, the connection is also created, using the connect built-in action:

create Networking_Client {
do connect to: "localhost" protocol: "tcp_client" port: 3001 with_name: "Client";
}

Each protocol has its specificities regarding the connection:

  • TCP:
    • protocol: the 2 possibles keywords are tcp_server or tcp_client, depending on the wanted role of the agent in the communication.
    • port: traditionally the port 3001 is used.
    • raw: false by default for compatibility purposes, however it is highly recommended to turn it to true when communicating with external applications as it will remove all the wrapper information used for communication inside gama and prevent some bugs when communicating inside gama.
  • Websocket:
    • protocol: the 2 possibles keywords are websocket_server or websocket_client, depending on the wanted role of the agent in the communication.
    • port: traditionally the port 3001 is used.
    • raw: false by default, it is better to turn it to true when communicating with external applications as it will remove all the wrapper informations used for communication inside gama.
  • UDP:
    • protocol: the 2 possibles keywords are udp_server or udp_emitter, depending on the wanted role of the agent in the communication.
    • port: traditionally the port 9876 is used.
  • MQTT:
    • protocol: MQTT is the default protocol value (if no value is given, MQTT will be used)
    • port: traditionally the port 1883 is used (when ActiveMQ is used as the server application)
    • admin and password: traditionally the default login and password are "admin" (when ActiveMQ is used as the server application)
  • HTTP requests:
    • protocol: the only keyword to use is http.
    • port: traditionally the port 80 is used for http connections and 443 for https.

Note: if no connection information is provided with the MQTT protocol (no port), then GAMA connects to an MQTT server provided by the GAMA community (for test purpose only!).

Sending messages

To send any message, the agent has to use the send action:

do send to: "server" contents: name + " " + cycle + " sent to server";

The network skill in GAMA allows the modeler to send simple string messages between agents but also to send more complex objects (and in particular agents). In this case, the use of the MQTT protocol is highly recommended.

do send to: "receiver" contents: (9 among NetworkingAgent);	

Receiving messages

Asynchronous reading

The messages sent by other agents are received in the mailbox attribute of each agent. So to get its new message, the agent has simply to check whether it has a new message (with action has_more_message() ) and fetch it (that gets it and remove it from the mailing box) with the action fetch_message().

reflex fetch when: has_more_message() {	
message mess <- fetch_message();
write name + " fecth this message: " + mess.contents;
}

Note that when an agent is received, the fetch of the message will recreate the agent in the current simulation.

Alternatively, the mailbox attribute can be directly accessed (notice that the mailbox is a list of messages):

reflex receive {  
if (length(mailbox) > 0) {
write mailbox;
}
}

Synchronous reading

In certain cases you need to wait for a message from another application to continue the execution of your simulation. To do so, you can use the fetch_message_from_network action to force the mailbox to refresh (which normally is only done once per cycle) until you receive a message:

reflex fetch {
write "waiting for server to send data";
loop while: !has_more_message() {
do fetch_message_from_network;
}

//This second loop will only be reached once a message has been found into the agent's mailbox
loop while: has_more_message() {
message s <- fetch_message();
write "at cycle: " + cycle + ", received from server: " + s.contents;
}
}

Broadcasting a message to all the agents' members of a given group

Each time an agent creates a connection to another agent as a client, a way to communicate with it is stored in the network_groups attribute. So an agent can use this attribute to broadcast messages to all the agents with whose it can communicate:

reflex broad {
loop id over: network_groups {
do send to: id contents: "I am Server " + name + " I give order to " + id;
}
}

To go further: