Home Open Problems Presented in RTSOPS

Open Problems Presented in RTSOPS

This is a running collection of open problems and challenges presented at the RTSOPS workshop (co-located with ECRTS). If you have updates to some of the open problems or would like to provide further information/details, you are welcome to reach out to the ECRTS organizers. Please note that LLMs were used to help creating the initial version of this page (for example by providing summaries from talk abstracts), so not all information may be entirely accurate or complete.

Open Challenges

In the following, open challenges of the RTSOPS Workshop are listed.

> Automotive Cause-Effect Chains in the Wild: A Real-World Industrial Benchmark and Generator

The real-time community lacks openly shareable, industrially representative cause-effect-chain workloads, which risks a widening gap between academic analysis and automotive practice — so the paper releases an anonymised statistical fingerprint of a real TTTech ADAS controller (102 tasks on a three-CPU/nine-core platform, 39 ASIL-tagged chains), plus a generator and 500 scalable synthetic instances that preserve automotive reality without exposing proprietary IP. Running seven public end-to-end-latency analyzers on this realistic data exposes the open problem the benchmark is meant to drive: they certify wildly different fractions of chains within budget (from just 9.8% up to 81.3%), so tighter chain analyses — and the synthesis of enforceable, jointly schedulable job-level dependencies — remain open.

Open Problems With Updates in the Past 5 Years (2022-2026)

This is an overview of the open problems posed at RTSOPS with updates in the past 5 years. Each entry gives a short description of the problem (if provided) followed by the events that have occurred around it (where it was stated, any follow-up talks, or papers that address the open problem).

> Reshaping Real-Time Workload Geometry to the Dimensions of Modern Hardware

  • Stated at RTSOPS 2026 by Bryan Ward (Vanderbilt University, USA)
  • Status: OPEN

> Worst-case memory latency and distributed DRAM banks

Memory latency is a critical component of worst-case timing, but modern COTS systems use multiple cache layers, physically distinct DRAM banks, and address-hashing to spread requests across banks for average-case parallelism — mechanisms that are hard to capture in worst-case analysis and that a trivial-but-safe bound renders overly pessimistic. The talk frames the problem using the Network Calculus formalism and presents improvements to it that capture some of these effects in order to reduce the pessimism gap.

  • Stated at RTSOPS 2026 by Raffaele Zippo (University of Pisa, Italy)
  • Status: OPEN

> Schedules, Lag, Fraenkel, and Tribonacci, all at Once

  • Stated at RTSOPS 2026 by Enrico Bini (University of Turin, Italy)
  • Status: OPEN

> Closing the Control Abstraction Gap: From Linear Control Models to Fine-Grained Verification of Control Software

While control design uses physical plant models at multiple levels of abstraction (from detailed nonlinear down to the simplified linear models used for synthesis), the software implementing the controller is usually captured only through coarse timing assumptions or high-level failure models — an asymmetry that leaves a gap between the controller, its implementing software, and the system to be verified. The open problem is to extend multi-granularity modelling to the controller’s software and timing behaviour, verifying implemented control software across several levels of fidelity (delays, missed updates, unfinished computations, synchronisation errors, and other code-level timing effects) through a principled framework that relates plant-model granularity to software-model granularity.

  • Stated at RTSOPS 2026 by Martina Maggio (Saarland University, Germany)
  • Status: OPEN

> Toward the Right Analytical Model and System Software for Autonomous Driving Systems: Open Problems and Research Directions

In autonomous driving systems, which turn multi-rate, asynchronous sensor streams into actuation through graphs of callbacks, nodes, and middleware, temporal correctness cannot be captured by any single task’s execution time or deadline — raising a two-sided question of what analytical models are needed to reason about AD timing and what system software (e.g. Autoware, ROS 2) is needed to realise, observe, and enforce them. The open problem is to close the gaps along five dimensions — task boundaries, timing metrics, resource models, execution-time variability, and safety integration — so that analytical models move closer to AD reality while AD system software is reshaped into analyzable, enforceable, and safety-aware infrastructure.

  • Stated at RTSOPS 2026 by Atsushi Yano and Takuya Azumi (Saitama University / TIER IV, Japan) [Extended Abstract]
  • Status: OPEN

> Challenges in the design of cause-effect chains: How to tune the end-to-end latency?

  • Stated at RTSOPS 2026 by Matthias Becker (KTH Royal Institute of Technology, Sweden)
  • Status: OPEN

> Expected end-to-end latency of cause-effect chains

Research on end-to-end latency of cause-effect chains focuses on bounding the maximum reaction time and maximum data age, but when combined with control-theoretic design, considering only the worst case yields suboptimal behaviour under normal operating conditions. The open problem is to derive probability distributions of end-to-end latency under uncertainty from jitter and deadline misses, enabling controllers that optimise the expected performance of cyber-physical systems while still guaranteeing worst-case compliance.

  • Stated at RTSOPS 2026 by Yde Sinnema (Lund University, Sweden) [Extended Abstract]
  • Status: OPEN

> Period Optimization and Priority Assignment for Cause-Effect Chains

Cause-effect chains describe how data propagates through a sequence of tasks, with timing measured by metrics such as end-to-end latency and data age. The problem concerns jointly choosing task periods and assigning priorities so that the timing behavior of these chains is optimized.

  • Stated at RTSOPS 2025 by Jian-Jia Chen and Mario Günzel.
  • Status: OPEN

> Can we design a model to automatically identify scheduling anomalies?

  • Stated at RTSOPS 2025 by Blandine Djika Mezatio, Alain Plantec, Georges Edouard Kouamou and Frank Singhoff.
  • Status: OPEN

> Towards Cycle Accuracy of Measurements of Software Running on an FPGA

  • Stated at RTSOPS 2025 by Caspar Treijtel.
  • Status: OPEN

> Self-Suspension Strikes Back

Self-suspension — when a task voluntarily pauses its execution — has been studied since 1988, yet across the different release constraints (frame-based, periodic harmonic, periodic arbitrary, sporadic) and suspension models (segmented, dynamic) two fundamental questions remain essentially unanswered: what is the computational complexity of scheduling self-suspending task sets, and when are such task sets schedulable? Current solutions to both are tightly bound to their specific task model, so that even a minor modification of the model reopens the questions.

> Exploring Partitioned and Semi-partitioned Callback Scheduling on ROS 2 Multi-threaded Executors

ROS 2 multi-threaded executors currently dispatch callbacks using a global, work-conserving policy, but the real-time community continues to debate whether partitioned or semi-partitioned scheduling would give better predictability. The authors add a thread-affinity API to rclcpp so callbacks can be pinned to specific executor threads, and show preliminary response-time benefits. The open problem is designing callback-to-thread allocation algorithms that account for callback utilisation and execution times, given ROS 2’s publisher-subscriber, event-triggered structure.

  • Stated at RTSOPS 2024 by Hoora Sobhani, Daniel Enright, Tejas Milind Deshpande, and Hyoseung Kim (University of California, Riverside, USA). [Abstract]
  • Status: OPEN

> Is there any optimal fixed-priority scheduling algorithm for probabilistic dependent tasks on one processor?

The setting is a set of real-time tasks on a single processor, related by directed-acyclic-graph (DAG) precedence constraints, where execution times and inter-arrival times are given by probability distributions and scheduling is preemptive fixed-priority. A priority assignment must keep the schedule stable (no response-time distribution diverging to infinity) while satisfying the precedence constraints. The open problem is whether there exists an optimal fixed-priority algorithm — one guaranteed to find a feasible priority assignment whenever one exists.

  • Stated at RTSOPS 2024 by Ismail Hawila (StatInf and Inria, France), Liliana Cucu-Grosjean (Inria, France), and Slim Ben Amor (StatInf, France). [Abstract]
  • Status: OPEN

> Adaptive Execution for Real-Time Observations of Astrophysical Transients

The goal is to minimise the expected latency from space-based detection of a transient event (e.g. a gamma-ray burst) to optical follow-up observation, trading off how long localisation takes against how large a sky region the follow-up telescope must then search. This spans several real-time open problems: deploying neural-network “slimming” and “early exit” together on an FPGA; quantifying how extra localisation compute shrinks the search region (not just the localisation error); and extending Buttazzo’s elastic scheduling model to independent jobs, including the competitive ratio of an online workload-compression algorithm against a clairvoyant one.

> New challenges in adaptive real-time systems with parametric WCET

Parametric WCET analysis computes a worst-case execution time as a formula over software/hardware parameters — instantiated once the parameters are known — instead of a single, usually pessimistic, number, which opens the door to adaptive systems. Building scheduling theory on top of such WCET formulas raises two open problems. First, sensitivity analysis: which input values make a task set schedulable, made hard by a single input affecting the WCET of several tasks. Second, semi-clairvoyant scheduling of tasks whose WCET is a formula, made hard by there being no fixed number of distinct WCET values per task.

  • Stated at RTSOPS 2023 by Clément Ballabriga, Julien Forget, Sandro Grebant, and Giuseppe Lipari (Univ. Lille, CNRS, Inria, Centrale Lille — CRIStAL, France). [Slides]
  • Status: OPEN

> Reachability-based response-time analysis: motivation, challenges, and open problems

  • Stated at RTSOPS 2022 by Mitra Nasri (Eindhoven University of Technology, TU/e). [Program entry]
  • Status: OPEN

> Extension of Multicore Response Time Analysis to analyse the Utilization of DMA Units

Multicore Response Time Analysis (MRTA) bounds worst-case response times while accounting for shared-resource interference, but it assumes each task runs on a single execution unit and cannot model offloading part of a task’s work to a DMA unit. The authors propose modelling the DMA in two roles — as a bus slave during configuration and a bus master during transfer — and splitting a task into precedence-constrained sub-tasks, each running on one hardware component. The open research questions are whether this master/slave modelling is appropriate, whether it generalises to other hardware accelerators, and whether the extended task model faithfully reflects the real timing behaviour.

  • Stated at RTSOPS 2022 by Alexander Stegmeier and Sebastian Altmeyer (University of Augsburg, Germany). [Paper]
  • Status: OPEN

> Resource allocation for complex DAG tasks with probabilistic execution times

The problem is how to allocate the sub-tasks of a DAG to resources (e.g. processor cores) so as to maximise a revenue function tied to a soft deadline on the end-to-end response time, where each sub-task’s execution time is a probability distribution (pWCET). Naive enumeration of allocations does not scale, so heuristics are required. The paper lists open issues including handling recurrent sub-tasks (in both response-time evaluation and allocation), imposing soft deadlines on individual sub-tasks, identifying which tasks are critical when a deadline is missed, and rescheduling the allocation when delay accumulates.

  • Stated at RTSOPS 2022 by Paola Cappanera, Laura Carnevali, Leonardo Paroli, Riccardo Reali, and Enrico Vicario (University of Florence, Italy). [Paper]
  • Status: OPEN

Open Problems Without Updates in the Past 5 Years

Open problems are moved down here if there has been no update in the past 5 years. (If there is no update within the past 5 years, problems will be migrated to here.)

To be migrated …

Editions of RTSOPS Before 2022

For completeness, these are the links to editions of RTSOPS before 2022. Open problems can be found in the program within each page.

Comments are closed.