Eri kriitilisusega ülesannete planeerimise meeetod
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Traditsiooniline lähenemine planeerimisele võib jätta süsteemid alakoormatuks, kuna arvestavad kõikide ülesannete puhul ainult maksimaalset ülesande täitmisele kuluvat aega. Reaalsetes süsteemides on aga ülesannete täitmisajad üldjuhul märgatavalt lühemad ning paljudel ülesannetel võivad olla ka mitteranged piirajad. Seetõttu peaksid planeerimisalgoritmid selliste omapäradega arvestama ning käesolevas magistritöös käsitletaksegi erineva kriitilisusega ülesannete planeerimist. Peamine erinevus seisneb mitte-kriitiliste ülesannete optimistlikumas planeerimises. Kasutusele on võetud erinevad prioriteetsusfunktsioonid alakoormatuse vähendamiseks ja põhitähelepanu on kiipvõrkudel põhinevatel kiipsüsteemidel. Esmalt uuritakse staatilist nimistu-põhist planeerimist.
Seejärel uuritakse prioriteetsusfunktsioonide mõju erineva kriitilisusega süsteemide planeerimisel. Käesolevas töös on ülesannete puhul kasutatud kahte erinevad kriitiisuse taset ja eesmärgiks on tagada kriitiliste ülesannete täitmine ettenähtud piiraegadeks ning mitte-kriitiliste ülesannete võimalikul optimaalne planeerimine. Ressursside vähesuse korral on võimalik ka mitte-kriitiliste ülesannete katkestamine või piiraegade mittesaavutamine. Kriitilised ülesanded seevastu täidetakse alati lõpuni ning ettenähtud piirajaks.
Traditional scheduling approaches waste utilization because they need to reserve absolute Worst Case Execution Time (WCET) for all tasks. However, in typical applications many tasks are executed much faster and have only soft deadlines. Such tasks are usually referred to as non-critical tasks. Hence, we have to take this into account while scheduling applications and in this thesis we introduce a scheduling approach that can schedule tasks with various criticalities. Apparently the predicted execution time for non-critical tasks should be less pessimistic than critical ones. Different priority functions may also be used to solve underutilization problem. In this thesis our concentration is on bringing mixed criticality into scheduling of the applications, executed on network-on-chip based systems. First of all we will illustrate an offline partitioning list scheduling. Later we explore the introduced mixed critical system with the impact of various priority functions and execution time models on the scheduling results. In our experiments we have used two levels of criticality and the goal is to guarantee the schedulability for the critical tasks and schedule the none-critical tasks in a best effort manner. In case the resources would be insufficient and a task would be possibly dropped or miss the deadline, the critical tasks are always guaranteed to meet their deadline.
Traditional scheduling approaches waste utilization because they need to reserve absolute Worst Case Execution Time (WCET) for all tasks. However, in typical applications many tasks are executed much faster and have only soft deadlines. Such tasks are usually referred to as non-critical tasks. Hence, we have to take this into account while scheduling applications and in this thesis we introduce a scheduling approach that can schedule tasks with various criticalities. Apparently the predicted execution time for non-critical tasks should be less pessimistic than critical ones. Different priority functions may also be used to solve underutilization problem. In this thesis our concentration is on bringing mixed criticality into scheduling of the applications, executed on network-on-chip based systems. First of all we will illustrate an offline partitioning list scheduling. Later we explore the introduced mixed critical system with the impact of various priority functions and execution time models on the scheduling results. In our experiments we have used two levels of criticality and the goal is to guarantee the schedulability for the critical tasks and schedule the none-critical tasks in a best effort manner. In case the resources would be insufficient and a task would be possibly dropped or miss the deadline, the critical tasks are always guaranteed to meet their deadline.