FairMQ/fairmq

FairMQ

The standard FairRoot is running all the different analysis tasks within one process. The FairMQ (Message Queue) allows starting tasks on different processes and provides the communication layer between these processes.

Devices

The components encapsulating the tasks are called devices and derive from the common base class FairMQDevice. FairMQ provides ready to use devices to organize the dataflow between the components (without touching the contents of a message), providing functionality like merging and splitting of the data stream (see subdirectory devices).

A number of devices to handle the data from the Tutorial3 detector of FairRoot are provided as an example and can be found in FairRoot/base/MQ directory. The implementation of the tasks run by these devices can be found FairRoot/example/Tutorial3. The implementation includes sending raw binary data as well as serializing the data with either Boost Serialization, Google Protocol Buffers or Root TMessage. Following the examples you can implement your own devices to transport arbitrary data.

Topology

Devices are arranged into topologies where each device has a defined number of data inputs and outputs.

Example of a simple FairMQ topology:

Topology configuration is currently happening via setup scripts. This is very rudimentary and a much more flexible system is now in development. For now, example setup scripts can be found in directory FairRoot/example/Tutorial3/ along with some additional documentation.

Messages

Devices transport data between each other in form of FairMQMessages. These can be filled with arbitrary content and transport either raw data or serialized data as described above. Message can be initialized in three different ways:

• with no parameters: This is usefull for receiving a message, since neither size nor contents are yet known.
• given message size: Initialize message body with a size and fill the contents later, either with memcpy or by writing directly into message memory.
• given message size and buffer: initialize the message given an existing buffer. This is a zero-copy operation.

After sending the message, the queueing system takes over control over the message body and will free it with free() after it is no longer used. A callback can be given to the message object, to be called instead of the destruction with free().

Transport Interface

The communication layer is available through an interface. Two interface implementations are currently available. Main implementation uses the ZeroMQ library. Alternative implementation relies on the nanomsg library. Here is an overview to give an idea how interface is implemented: