Multi-Rate Task Scheduling: 100% Schedulability Without the LET Latency Trap

If you're running a safety-critical control loop and using Logical Execution Time (LET) to guarantee determinism, you're injecting latency that degrades your data's age – but a new framework from arXiv:2603.09738 fixes that without sacrificing schedulability.

The LET Trade-off: Determinism at the Cost of Freshness

LET has been the go-to paradigm for compositional determinism in autonomous systems, but it comes with a hidden tax: buffered communication adds artificial latency. For high-frequency control loops — think 1 kHz or faster — that injected latency directly ages the data reaching the actuator, potentially destabilizing the system. Oversampling multi-rate task dependencies to compensate is inefficient and often fails under tight deadlines.

Backward Tracing from Actuators to Find Dominant Paths

The authors reformulate the problem by treating data freshness as a first-class scheduling constraint. They start by decomposing the Data Dependency Graph into dominant paths, tracing the strictest freshness constraints backward from the actuators. This isn't the usual forward mapping from sensors to controllers; it's a reverse search that pinpoints exactly which chain of tasks demands the youngest data.

Once the dominant paths are identified, an offset search algorithm synchronizes the start times of tasks across the multi-rate, multi-dependency chains. The result: end-to-end data freshness is enforced without the artificial delay that LET buffering introduces. The trade-off is a shift away from pure execution determinism toward a controlled balance between freshness and predictability.

Offset Alignment Preserves Global EDF Capacity

The key claim — and it's backed by a formal proof — is that this offset-based alignment preserves the 100% schedulability capacity of Global Earliest-Deadline-First (EDF). In practice, that means you don't lose any scheduling headroom by adding freshness guarantees. The algorithm is essentially a zero-cost constraint enforcement mechanism, at least from a schedulability standpoint.

For anyone building real-time systems for autonomous vehicles, drones, or industrial robots, this is a direct attack on a persistent pain point: how to keep control loops deterministic without starving them of fresh sensor data. The math is spelled out in the paper, and the proof of schedulability preservation means the approach is ready for integration into existing real-time OS schedulers.

Source: Ensuring Data Freshness in Multi-Rate Task Chains Scheduling
Domain: arxiv.org