[Paper Review] Sufficient FTP Schedulability Test for the Non-Cyclic Generalized Multiframe Task Model

Vandy Berten, Joël Goossens|arXiv (Cornell University)|Oct 26, 2011

Real-Time Systems Scheduling8 references3 citations

TL;DR

This paper presents a sufficient schedulability test for non-cyclic generalized multiframe (nc-gmf) tasks under fixed-priority scheduling (FTP), leveraging a critical instant analysis and a request-bound function-based algorithm. The key contribution is a pseudo-polynomial-time algorithm that guarantees feasibility by bounding workload using a maximal request-bound function (mrbf), with theoretical bounds on pessimism and potential for optimization via periodic pattern detection.

ABSTRACT

Our goal is to provide a sufficient schedulability test -ideally polynomial- for the scheduling of Non-Cyclic Generalized Multiframe Task Model using Fixed-Task-Priority schedulers. We report two first results: (i) we present and prove correct the critical instant for the Non-Cyclic Generalized Multiframe Task Model then (ii) we propose an algorithm which provides a sufficient (but pseudo-polynomial) schedulability test.

Motivation & Objective

• To address the lack of a sufficient schedulability test for non-cyclic generalized multiframe (nc-gmf) tasks under fixed-priority scheduling (FTP).
• To extend the critical instant concept to nc-gmf models, where task configurations are non-cyclic and not known in advance at design time.
• To develop a polynomially bounded, pseudo-polynomial schedulability test that ensures all jobs meet their deadlines under FTP.
• To quantify and bound the pessimism introduced by the fixed-priority model compared to dynamic-priority approaches.
• To explore future extensions to the digraph real-time task model and to improve algorithmic efficiency via periodic pattern detection in workload functions.

Proposed method

• Proposes a critical instant analysis showing that the worst-case response time for any nc-gmf task occurs when all higher-priority tasks release jobs synchronously and as early as possible.
• Introduces the maximal request-bound function (mrbf_i(t)) as a workload upper bound for each task τ_i, computed over all possible configuration sequences.
• Defines an offset-independent scenario model where job sequences are independent of initial release times, enabling worst-case analysis.
• Develops an algorithm to compute mrbf_i(t) by exploring all valid configuration sequences up to time t, using pruning via lower and upper workload bounds.
• Applies a pruning condition: if t′ × U_max < v′ + C_max and (t′, v′) is not in the queue, the point is ignored to reduce search space.
• Uses the line U_max × t as a lower bound (lb_i(t)) and U_max × t + (C_max − C_U_max) as a refined lower bound to guide the search and reduce pessimism.

Experimental results

Research questions

• RQ1 What is the worst-case scenario for response time in the non-cyclic generalized multiframe task model under fixed-priority scheduling?
• RQ2 Can a sufficient schedulability test be constructed for nc-gmf tasks under FTP that is both correct and computationally feasible?
• RQ3 How can the workload of nc-gmf tasks be bounded using a maximal request-bound function (mrbf) that accounts for arbitrary configuration sequences?
• RQ4 What are the sources of pessimism in the FTP-based schedulability test, and can they be quantified or reduced?
• RQ5 Can the periodic structure of mrbf_i(t) be exploited to improve the time complexity of the schedulability test?

Key findings

• The worst-case response time for any nc-gmf task occurs in a synchronous scenario where all higher-priority tasks release jobs simultaneously and as early as possible.
• The maximal request-bound function mrbf_i(t) is always above the line U_max × t, which serves as a valid lower bound for all possible workload scenarios.
• The algorithm for computing mrbf_i(t) is pseudo-polynomial in time and space, bounded by Δ and n_i, with pruning based on workload bounds to reduce search space.
• A point (t′, v′) can be safely ignored if t′ × U_max < v′ + C_max and (t′, v′) is not in the queue, significantly reducing computation.
• The workload function mrbf_i(t) eventually follows a periodic pattern of size C_i^{U_max} × T_i^{U_max}, suggesting potential for optimization via pattern detection.
• The method introduces pessimism compared to dynamic-priority models, but this pessimism can be quantified and potentially bounded through further refinement.