Erineva kriitilisusega ülesannete veakindel planeerimine mitmetuumaliste süsteemides
Date
2013
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Tartu Ülikool
Abstract
Käesoleva magistritöö eesmärgiks on erineva kriitilisusega rakenduste disainiprotsessi toetamine kiipvõrkudel põhinevate kiipsüsteemide süsteemitaseme disainivoos. Kiipvõrkudel põhinevad süsteemid on üks võimalikest platvormidest järgmise generatsiooni küber-füüsikalistes süsteemides. Samal ajal on suureneb vajadus erineva kriitilisusega rakenduste käitamiseks sellistes süsteemides. Seetõttu on vajalik süsteemitaseme töövahendite arendamine, et need toetaksid ka erineva kriitilisusega rakendusi. Käesolevas töös on laiendatud olemasolevat süsteemitaseme töökeskkonda, lisades sellele arendusvahendid erineva kriitilisusega arvestamiseks. Eesmärgiks on luua modelleerimis- mitte simuleerimiskeskkond, et tagada disainiprotsessi kiirus. Samal ajal on tagatud loodavate planeeringute täpsus, mis võimaldab nende kasutamist simulaatorites.
Käesolev magistritöö laiendab olemasolevat süsteemitaseme töökeskkonda. Peamiseks eesmärgiks on ressursside kasutamise optimeerimine, mis tekib rangete reaalajasüsteemide puhul. Samal ajal on eesmärgiks usaldusväärsuse ja süsteemi käitumise ennustatavuse säilitamine, et tagada kriitiliste komponentide ajaliselt korrektne käitumine. Loodud mudel võimaldab erineva kriitilisuse arvutamist nii ülesannetes kui ka kommunikatsioonis. Samuti on loodavad planeeringud võimelised tolereerima teatud arvu vigu nii ülesannetes kui ka kommunikatsioonis. Erineva kriitilisusega ülesandeid ja kommunikatsiooni käsitletakse erinevalt. Kui kriitiliste ülesannete puhul on alati eelduseks maksimaalne võimalik täitmisaeg, siis mitte-kriitiliste ülesannete puhul arvestatakse keskmist täitmisaega. Tulemused on tõestatud erinevate eksperimentidega ning nende põhjal võib väita, et erineva kriitilisuse arvestamine aitab märgatavalt tõsta süsteemide efektiivsust ja suurendada süsteemi koormatust, samas säilitades ajalise korrektsuse.
The goal of this thesis is to bring support of mixed-critical applications to system-level design flow of dependable Network-on-chip-based (NoC-based) multiprocessor systems. NoC-based multiprocessor technology is emerging as the next-generation platform for cyber-physical-systems. At the same time, there is a motivation to adopt mixed-critical applications in such systems. Therefore, there is a need for a suitable system-level design (SLD) framework which also supports development of mixed-critical systems. We choose a recent SLD framework for NoC-based systems and enhance it to support mixed-critical applications. This framework performs modeling rather than simulation which makes it much faster than a NoC simulator. At the same time it generates cycle accurate schedules to be verified against the simulators. However, this SLD framework does not support mixed-critical applications. We develop our mixed-critical solution on top of this SLD framework. We intend to deal with the problem of resource under-utilization which is associated with mixed-criticality and affects system performance. At the same time we need to maintain dependability and predictability of the critical parts of the application. In our solution, we develop an application model which supports mixed-criticality for both tasks and messages. Furthermore, we develop a mixed-critical fault-tolerance for computation and communication scheduling of our SLD framework. Our mixed-critical approach identifies tasks and messages of different criticality. In order to tackle the problem of under-utilization, the fault-tolerance in our approach is worst-case scenario for critical parts whereas it is average-case scenario for non-critical parts and it considers both design-time estimation and run-time condition in terms of fault occurrence. We have run several experiments on various applications and with different design settings in order to evaluate our solution. With the solution developed in this thesis, we have successfully enhanced our SLD flow to handle and schedule dependable mixed-critical applications. We have addressed the resource under-utilization problem and achieved performance improvements in the generated schedules.
The goal of this thesis is to bring support of mixed-critical applications to system-level design flow of dependable Network-on-chip-based (NoC-based) multiprocessor systems. NoC-based multiprocessor technology is emerging as the next-generation platform for cyber-physical-systems. At the same time, there is a motivation to adopt mixed-critical applications in such systems. Therefore, there is a need for a suitable system-level design (SLD) framework which also supports development of mixed-critical systems. We choose a recent SLD framework for NoC-based systems and enhance it to support mixed-critical applications. This framework performs modeling rather than simulation which makes it much faster than a NoC simulator. At the same time it generates cycle accurate schedules to be verified against the simulators. However, this SLD framework does not support mixed-critical applications. We develop our mixed-critical solution on top of this SLD framework. We intend to deal with the problem of resource under-utilization which is associated with mixed-criticality and affects system performance. At the same time we need to maintain dependability and predictability of the critical parts of the application. In our solution, we develop an application model which supports mixed-criticality for both tasks and messages. Furthermore, we develop a mixed-critical fault-tolerance for computation and communication scheduling of our SLD framework. Our mixed-critical approach identifies tasks and messages of different criticality. In order to tackle the problem of under-utilization, the fault-tolerance in our approach is worst-case scenario for critical parts whereas it is average-case scenario for non-critical parts and it considers both design-time estimation and run-time condition in terms of fault occurrence. We have run several experiments on various applications and with different design settings in order to evaluate our solution. With the solution developed in this thesis, we have successfully enhanced our SLD flow to handle and schedule dependable mixed-critical applications. We have addressed the resource under-utilization problem and achieved performance improvements in the generated schedules.