Performance Evaluation of Scheduling Tasks in Many-Core Systems Using Processes and Threads

Jul 16, 2026

Upload by Violet

This study evaluates scalability of process-based and thread-based schedulers for many-core shared-memory systems using a memory-intensive row-wise quick-sort workload on large 3D tensors. Process scheduling compares bounded prolific, bounded collective, and three pipe-based producer–consumer designs (one-to-one, one-to-many, many-to-many) with streaming task identifiers to workers to improve runtime load balancing. Thread scheduling analyzes static, dynamic, guided, chunk-based, chunkstealing, adaptive chunk, and AIMD strategies, where AIMD uses TCP-like additive-increase/multiplicative-decrease with EWMA CPU utilization to control a contention window. Experiments on a 24-core x86-64 platform show thread schedulers achieve the highest overall performance, with dynamic and guided approaches performing best, while pipe-based process designs scale strongly, selecting one-to-one for smaller workloads and many-to-many for larger ones.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

Introduction

Process-based scheduling

Thread scheduling

Scalability evaluation

Performance evaluation of scheduling tasks in many-core systems utilizing processes and threads arXiv :2607 .0482 1v 1 [ cs .DC] 6 Jul 2026 Mejgan Dedaj MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [dedajmegi@gmail.com](dedajmegi@gmail.com) Theofanis Ioannou MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [th.ioannou01@gmail.com](th.ioannou01@gmail.com) Hermione Kimpouropoulou MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [hermionekimp@gmail.com](hermionekimp@gmail.com) Kleopatra Kontogianni MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [kontkleo13@yahoo.gr](kontkleo13@yahoo.gr) Michail Panagiotidis Kannas MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [kannasmichael@gmail.com](kannasmichael@gmail.com) Anna Maria Sidiropoulou MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [annasidi360@gmail.com](annasidi360@gmail.com) Argyro Gailla MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [silvigail@hotmail.com](silvigail@hotmail.com) Stamatia Kastrinaki MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [matinakst@gmail.com](matinakst@gmail.com) Dimitrios Kontodimos MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [Jimkon03@gmail.com](Jimkon03@gmail.com) Sotirios Kontogiannis* MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [skontog@uoi.gr](skontog@uoi.gr) Anastasia Papouda MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [natasapapouda@gmail.com](natasapapouda@gmail.com) George Tavridis MicroComputer Systems Laboratory Dept. of Mathematics, University of Ioannina [georgetavridis7@gmail.com](georgetavridis7@gmail.com) Abstract This study assesses the scalability of process-based and thread-based schedulers for many-core shared-memory systems using a memory-intensive row-wise quick-sort workload on large three-dimensional tensors. The process-based evaluation considers bounded prolific, bounded collective, and three pipe-based producer-consumer schedulers: oneto-one, one-to-many, and many-to-many. These pipe schedulers dynamically stream task identifiers to worker processes, exchanging increased inter-process communication These authors contributed equally to this work. * Correspondence to: [skontog@uoi.gr](skontog@uoi.gr) overhead for enhanced runtime load balancing and flexible chunk-based task dispatching. The thread-based evaluation examines static, dynamic, guided, chunk-based, chunkstealing, adaptive chunk, and AIMD adaptive scheduling strategies. The AIMD scheduler employs an additiveincrease multiplicative-decrease policy inspired by TCP congestion control, utilizing an exponentially weighted moving average (EWMA) of CPU utilization to regulate a contention window that limits the number of concurrently active chunks. The adaptive chunk scheduler further modifies chunk size based on observed per-thread execution speed. Experimental results on a 24-core x86- 64 platform indicate that thread schedulers deliver the highest overall performance, with dynamic and guided scheduling yielding the most favorable practical outcomes. Among process schedulers, pipe-based designs demonstrate the strongest scalability, with one-to-one pipes excelling for smaller workloads and many-to-many pipes preferred for larger workloads. In summary, lightweight thread scheduling is optimal for shared-memory row sorting, while AIMD/adaptive scheduling and pipe-based process scheduling remain valuable for contention-aware execution, explicit inter-process coordination, and distributedstyle heterogeneous workload management. 1. Introduction Process-based scheduling is a central perspective for heavyweight tasks executed in a Single Instruction Multiple Process (SIM

FAQs

What workload and system model are used to evaluate scheduling performance?

The evaluation uses a memory-intensive row-wise quick-sort workload on large three-dimensional tensors in many-core shared-memory systems.

Which thread scheduling strategies are compared in the study?

The thread-based evaluation includes static, dynamic, guided, chunk-based, chunkstealing, adaptive chunk, and AIMD adaptive scheduling strategies.

How do process-based pipe schedulers differ, and what scalability trends are reported?

Process pipe schedulers stream task identifiers using one-to-one, one-to-many, and many-to-many producer–consumer designs; one-to-one excels for smaller workloads, while many-to-many is preferred for larger workloads.

Jul 16, 2026

Upload by Violet