There are several co-simulation architectures that can be constructed where the relationships between the federates, cores, and brokers can vary.
Broker topology is somewhat optional for simple co-simulations, but offers an increase in performance if it is possible to identify groups of federates that interact often with each other but rarely with the rest of the federation. In such cases, assigning that group of federates their own broker will remove the congestion their messages cause with the federation as a whole. The Fundamental Topics and Examples are built with a single broker.
The figure below shows the most common architecture for HELICS co-simulation. Each core has only one federate as an integrated executable, all executables reside on the same computer and are connected to the same broker. This architecture is particularly common for small federates and/or co-simulations under development. This is also the architecture for the Fundamental Examples.
Common HELICS architecture
Multiple Federates on a Single Core¶
The architecture below shows a scenario where more than one federate is associated with a single core. For most simulators that have already been integrated with HELICS this architecture will generally not be used. For simulators that are multi-threaded by nature and typically represent multiple independent simulated entities (federates to HELICS), HELICS can be configured to facilitate message passing between threads. For a co-simulation that exists entirely within a single executable, this architecture will provide the highest performance. For example, if a large number of small controllers are written as a single, multi-threaded application (perhaps all the thermostats in a large commercial building are being managed by a centralized controller), particularly where there is communication between the federates, using a single core inside a single multi-threaded application (with typically one thread per federate) will provide the highest level of performance.
Multi-threading allowing multiple federates on a single HELICS core
Computationally Heavy Federates¶
For co-simulations on limited hardware where a federate requires significant computational resources and high performance is important, it may be necessary to spread the federates out across a number of compute nodes to give each federate the resources it needs. All federates are still connected to a common broker and it would be required that the computers have a valid network connection so all federates can communicate with this broker. In this case, it may or may not be necessary to place the broker on its own compute node, based on the degree of competition for resources on its current compute node.
Architecture for multiple compute nodes
Alternatively, it would be possible to locate a broker on each computer and create a root broker on a third node. This kind of architecture could help if higher performance is needed and the federates on each computer primarily interact with each other and very little with the federates on the other computer. As compared to the previous architecture, adding the extra layer of brokers would keep local messages on the same compute node and reduce congestion on the root broker. An overview of how this is constructed is provided in the section on broker hierarchies and an example of this architecture (though running on a single compute node for demonstration purposes), is shown in the broker hierarchy example
Multiple broker architecture