Reducing Integration Costs and Fostering Collaboration in Distributed Training Systems with Sen: an Open Source Software Framework by Airbus
We examine gaps observed when specialised products are integrated into larger systems, their causes, and explore a mechanism addressing part of the recurring cost. Integration cost in distributed training systems is structurally hard to observe. Projects cannot see assets held by other projects, partners are often also competitors, solutions are built without regard to the integration cost the customer pays, and turnover resets what is known. The same structure that hides the cost makes it recur, at every scale from team to company and across the industry, so figures are scarce and the problem persists. At Airbus, we see how this tends to isolate developments that could otherwise naturally fit into tool ecosystems, where synergies would open technological advances while reducing costs for all parties.
Method: not a controlled comparison but a longitudinal observation: twenty years in defence simulation and automotive programmes, and the design and operation of three successive infrastructures in one division. That position makes recurrence visible. Single-project accounting does not. It also carries the biases of a participant rather than an observer. The distinction between interoperability and composability is not new; it has been discussed in literature and is generally well understood.
This paper addresses the engineering realisation of assembly. HLA and DDS decouple where a component runs from what it does, but neither provides a local composable programming model, so co-located components still pay to reach each other.
Mechanism: Sen is a general-purpose distributed object-oriented integration framework, released by Airbus as FOSS under the permissive Apache 2.0 license, and under active development. It lets independently built tools be connected into ecosystems its users define, so integration effort accrues to the community instead of being repaid at each boundary. Sen separates a component's placement from the cost of reaching it. A mechanism not aimed at making money, but at saving costs and fostering innovation, with no vendor-lock constraints.
Components are written once against a system model, then placed in a thread, process, or machine by configuration alone. Co-located components do not pay transport costs. Environments can be reconfigured as needs change rather than rebuilt. Sen is broker-less, with compile-time and run-time introspection, and supports real-time, faster-than-realtime, and stepped execution. Sen assumes a protected enclave, leaving security to the deployment environment, and targets non-safety-critical use.
What was observed: Sen is the integration substrate of a whole engineering domain, across ten programmes with three simulators in service, with several external suppliers using it where their systems meet Airbus systems. Its predecessor, on the same principle, displaced licensed middleware over sixteen years and made work viable that had not been affordable before. Interfaces drift, and that is where integration effort goes; automatic conversion between versions, lossless where possible and explicitly lossy where not, lets organisations on different release cycles work together without lockstep upgrades. Switching cost, not technical merit, often decides whether shared infrastructure spreads: it is taken up where two parties must meet and rarely elsewhere, which is why reuse mandates written into requirements so often fail.