[Paper Review] Scheduling of Hard Real-Time Multi-Thread Periodic Tasks

Irina Iulia Lupu, Joël Goossens|arXiv (Cornell University)|May 25, 2011

Real-Time Systems Scheduling14 references3 citations

TL;DR

This paper proposes a novel parallel task model for hard real-time systems where periodic tasks consist of multiple independent threads that can execute simultaneously on different processors. It introduces hierarchical and global thread schedulers with exact schedulability tests, demonstrating through simulations that multi-thread scheduling outperforms Gang scheduling in success ratio and response time, especially as processor count increases.

ABSTRACT

In this paper we study the scheduling of parallel and real-time recurrent tasks. Firstly, we propose a new parallel task model which allows recurrent tasks to be composed of several threads, each thread requires a single processor for execution and can be scheduled simultaneously. Secondly, we define several kinds of real-time schedulers that can be applied to our parallel task model. We distinguish between two scheduling classes: hierarchical schedulers and global thread schedulers. We present and prove correct an exact schedulability test for each class. Lastly, we also evaluate the performance of our scheduling paradigm in comparison with Gang scheduling by means of simulations.

Motivation & Objective

• To address the lack of realistic task models and schedulability analysis for periodic, multi-threaded real-time applications on identical multiprocessor platforms.
• To design a new parallel task model allowing independent threads of a task to execute concurrently across multiple processors.
• To develop predictable schedulers—hierarchical and global thread schedulers—based on Fixed Task Priority (FTP) and Fixed Subprogram Priority (FSP).
• To provide exact schedulability tests for both scheduler classes, ensuring correctness and feasibility analysis for real-time systems.
• To empirically evaluate the performance of the proposed model against Gang scheduling, particularly in terms of success ratio and worst-case response time.

Proposed method

• Introduces a formal task model where each periodic task τi is defined by an offset Oi, a set of v_i subprograms (each with individual worst-case execution time C_i^j), a period Ti, and a relative deadline Di.
• Distinguishes between tasks (off-line definitions) and processes/threads (run-time instances), with threads scheduled independently on available processors.
• Classifies schedulers into two classes: hierarchical (assigning priorities at task and subprogram levels) and global thread schedulers (assigning thread priorities independently of task structure).
• Develops exact schedulability tests based on the predictability of the scheduling behavior and the periodic nature of the tasks, leveraging known results from real-time theory.
• Employs simulation-based evaluation using synthetic workloads to compare success ratio and worst-case response time (WCRT) between the proposed multi-thread scheduler and Gang scheduling.
• Extends the model to multi-phase tasks (sequential phases with varying degrees of parallelism) to assess practicality, revealing non-predictability in hierarchical schedulers.

Experimental results

Research questions

• RQ1 Can a realistic parallel task model be defined that allows independent threads of a periodic task to execute simultaneously on multiple processors?
• RQ2 Are hierarchical and global thread schedulers based on FTP and FSP predictable in the context of multi-thread periodic tasks?
• RQ3 Can exact schedulability tests be formally derived and proven correct for these scheduler classes?
• RQ4 How does the performance of the proposed multi-thread scheduling approach compare to Gang scheduling in terms of success ratio and response time?
• RQ5 What are the limitations of extending the model to multi-phase tasks with sequential and parallel execution phases?

Key findings

• The proposed multi-thread scheduling approach achieves a success ratio up to 7.5 times higher than Gang scheduling on 16-processor platforms at utilization 10.4.
• On 4, 8, and 16 processors, the (DM, IM) scheduler reduces worst-case response time (WCRT) of the lowest-priority task by up to 8% compared to Gang DM, with performance gaps increasing with processor count.
• For 2-processor platforms, the WCRT of the lowest-priority task is equal under both (DM, IM) and Gang DM, but performance advantage emerges on larger platforms.
• The (DM, IM) scheduler outperforms Gang DM in at most 50% of the tested task systems in terms of WCRT, with the majority showing superior response time.
• The proposed schedulers are predictable and allow exact schedulability tests, unlike previous approaches that rely on heuristics or sufficient conditions.
• Extending the model to multi-phase tasks (e.g., initialization, parallel computation, finalization) results in non-predictable behavior for hierarchical schedulers, as demonstrated by a counterexample with timing anomalies.