Context.In the context of Software Architecture, AADL (ArchitectureAnalysis and Design Language) is one of the latest standards (SAE StandardAS5506) used for analyzing and designing of architectures of software sys-tems. Dependability evaluation of an avionic system, modeled in AADL, isconducted using petri nets standard PNML (ISO standard ISO/IEC15909-2).A maintainability analysis of PNML dependability model is also con-ducted. Objectives. In this study we investigate maintainability analysis of PNMLdependability model of an avionic system designed in AADL. Structural,functional, fault-tolerance and recovery dependencies are modeled, imple-mented, simulated and validated in PNML. Maintainability analysis withrespect to ‘changeability’ factor is also conducted. Methods.This study is a semi-combination of ’case-study’ and ’implemen-tation’ research methodologies. The implementation of case-study systemis conducted by modeling the case-study system in AADL using OSATE2tool and simulating the dependability models in PNML using Wolfgangtool. PNML dependability models are validated by comparing with GSPNdependability models of previously published research. Results. As a result of this research, PNML dependability model was ob-tained. The difficulties that influenced the research in AADL Error ModelAnnex and the OSATE2 tool are also analyzed and documented. PNMLand GSPN are compared for complexity. And maintainability analysis forPNML dependability model w.r.t ‘changeability’ factor is also an outcomeof this research. This research is recommended for software testing at ar-chitecture level as a standardized way for testing the software componentsfor faults and errors and their impact on dependable components. Conclusions. We conclude that PNML is an ISO standard and is the al-ternative for GSPN for dependability. Also, AADL Error Model Annex isstill evolving and there is a need of availability of proper literature publiclyfor better understanding. Also, PNML dependability model possesses the‘changeability’ factor of maintainability analysis and therefore it is able toadapt changes in the architecture. Also, dependability factors of a softwarecan be tested at architecture level using the standards; AADL and PNML

The expansion and increased complexity of technology is undoubtedly consistent and one can intuitively suppose that this trajectory will not deviate from this course in the years to come. On a continuous basis, concepts that started of as some hypothetical or abstract notions without practical relevance gets transferred to the modern state of our current technology. During these times, where a subset of our technology has the responsibility of handling the safety of our being, research within dependability theory must keep up the pace with technology. One cannot emphasize enough the importance of ensuring the validity of system dependability attributes prior and posterior to development. With the objective of aggregating findings to the research field and potentially derive new propositions this paper assesses the stochastic modeling concepts used within dependability theory. In particular, discrete-time-and continuous-time-Markov chains are outlined in detail, searching for possibilities to extend these processes in the context of real-time system monitoring. As an outcome, numerical 'repair rate optimization' (RRO) through CTMC uniformization is introduced. A technique which deduces a proposed allocation of repair rate adjustment based on the models parametric sensitivities (gradient ascent). The theoretical aspects are verified by development of an algorithm in Matlab that utilizes the above. Additionally, an approach of combining dependability attributes into a unified measure is proposed. Where the (bounded) transient probabilities are regarded as vectors in the L2(R, B(R), λ) Hilbert Space. For which a normalized dependability norm can be obtained by using the induced norm and triangle inequality. This serves as a metric to compare distinct architectures in terms of several, quantitative attributes. The results imply that under the hypothesis that the system/company can adapt to an increased demand on maintenance periodicity, reliability/availability can be significantly improved. Mitigating risk of failure while optimally preserving resources in terms of e.g. core capacity, maintenance personnel, budget and/or required redundancy while conditioning on the actual system behaviour.

Due to the usage of complicated platforms and current high-performance space computing technology, onboard processing in small satellites is expanding. Space-cloud payload processors with Commercial Off-The-Shelf (COTS) components, that are required to be radiation-tolerant, are used to perform the onboard processing. In this thesis, the research will aim to increase the dependability of a generic space-cloud payload processor through its Solid State Drive (SSD) storage unit. To achieve this, a more dependable NAND-flash-based SSD Redundant Array of Independent Disc (RAID) storage system is designed and tested. The reliability of NAND-flash-based SSDs can suffer wear-outs due to increased Program/Erase (P/E) cycles, making them more prone to radiation effects. These radiation effects are considered non-destructive events in the form of bit errors (both single bit-flip and multiple bit-flips). Therefore, making the storage system more dependable involves increasing its reliability against non-destructive events and developing analytical models that account for the considered dynamic of the SSD RAID. The challenge that comes with achieving the aim of this thesis is twofold. First, to explore different RAID levels such that a combination of RAID levels can be incorporated into one SSD for better reliability than a RAID-1 setup. Hence, in this thesis, a RAID array of several SSDs is not considered. Furthermore, the combinations of RAID levels need to account for mixed-critical data. Second, to demonstrate, via simulation and analytical models, the impact on the reliability of the storage system. A comparison study is also undertaken due to the support that the Fourth Extended (Ext4) file system or Zettabyte File System (ZFS) may give to enhance the storage system, and since little research exists that compares the file systems in some feature categories. The solution is a RAID-5 + 6 storage system that is Error Detection And Correction (EDAC) protected by Hamming codes and Reed Solomon (RS) codes. Low-critical data is stored using RAID-5 whereas high-critical data is stored using RAID-6. The simulation of the storage system proves that low-critical stripes of data achieve single fault tolerance whereas high-critical stripes of data tolerate a maximum of 5-bit burst errors. In parallel, several Continuous Time Markov Chain (CTMC) models are analysed, which show that the proposed solution is indeed highly reliable. The comparison study is carried out in a systematic way, and the findings are established as substantial,i.e., ZFS provides greater storage system support. In summary, the results of creating the storage system and analysing it suggest that incorporating RAID-5 and RAID-6 offers better SSD RAID reliability than RAID-1. / Användningen av komplicerade plattformar och aktuell högpresterande rymdberäkningsteknik expanderar onboard-processing i små satelliter. Space-Cloud lösningar med kommersiellt tillgängliga komponenter som är toleranta mot strålningar i rymden används för att utföra onboard-processing. I detta examensarbete syftar forskningen till att förbättra tillförlitligheten hos en generisk rymd dator genom dess SSD-lagringsenhet. För att uppnå detta har ett mer tillförlitligt lagringssystem bestående av NAND-flash och RAID designats och testats. Tillförlitligheten hos NAND-flash-baserade SSD:er kan försämras då dessa kan drabbas av slitage på grund av ökade P/E cykler, vilket gör dem mer benägna för strålningseffekter. Dessa strålningseffekter anses vara icke-destruktiva i form av bit-fel (både enskilda bit-flippar och flera bit-flippar). Med denna anledning görs lagringssystemet mer tillförlitligt för att tolerera icke-destruktiva händelser. Utöver detta, utvecklas analytiska modeller som tar hänsyn till den betraktade dynamiken i SSD RAID. Utmaningen som följer med att uppnå syftet med denna avhandling är tvådelad. För det första, för att utforska olika RAID-nivåer så att en kombination av RAID-nivåer kan inkorporeras i en SSD för bättre tillförlitlighet än RAID-1. Således övervägs inte en RAID-array av flera SSD:er i denna avhandling. Dessutom måste kombinationerna av RAID-nivåer ta hänsyn till data av olika kritikalitet. För det andra, för att genom simulering och analytiska modeller indikera påverkan på lagringssystemets tillförlitlighet. En jämförelsestudie genomförs också på grund av stödet som filsystemen Ext4 eller ZFS kan ge för att förbättra lagringssystemet och eftersom det finns lite forskning som jämför filsystemen i några funktionella kategorier. Lösningen baseras på ett RAID-5+6 lagringssystem som är skyddat av Hamming-koder och RS koder för att upptäcka fel och korrigera dem. Lågkritisk data lagras med RAID-5 medan högkritisk data lagras med RAID-6. Simuleringen av lagringssystemet visar att lågkritiska datasektioner uppnår en fel tolerans mot enskilda bit-flippar medan högkritiska datasektioner kan tåla maximalt 5 bit-flippar. Samtidigt analyseras flera CTMC modeller som visar att den föreslagna lösningen verkligen är mycket tillförlitlig. Jämförelsestudien utförs på ett systematiskt sätt och resultaten fastställs som betydande, det vill säga att ZFS ger större stöd för lagringssystemet. Sammanfattningsvis antyder resultaten av att skapa lagringssystemet och analysera det att inkorporering av RAID-5 och RAID-6 erbjuder bättre tillförlitlighet för SSD RAID än RAID-1.

Made available in DSpace on 2014-12-17T14:55:09Z (GMT). No. of bitstreams: 1 IvanovitchMDS_TESE_Capa_ate_pag78.pdf: 3947167 bytes, checksum: ce8af4212a09f9d5cf62d132879a91d3 (MD5) Previous issue date: 2013-01-21 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior / Ensuring the dependability requirements is essential for the industrial applications since faults may cause failures whose consequences result in economic losses, environmental damage or hurting people. Therefore, faced from the relevance of topic, this thesis proposes a methodology for the dependability evaluation of industrial wireless networks (WirelessHART, ISA100.11a, WIA-PA) on early design phase. However, the proposal can be easily adapted to maintenance and expansion stages of network. The proposal uses graph theory and fault tree formalism to create automatically an analytical model from a given wireless industrial network topology, where the dependability can be evaluated. The evaluation metrics supported are the reliability, availability, MTTF (mean time to failure), importance measures of devices, redundancy aspects and common cause failures. It must be emphasized that the proposal is independent of any tool to evaluate quantitatively the target metrics. However, due to validation issues it was used a tool widely accepted on academy for this purpose (SHARPE). In addition, an algorithm to generate the minimal cut sets, originally applied on graph theory, was adapted to fault tree formalism to guarantee the scalability of methodology in wireless industrial network environments (< 100 devices). Finally, the proposed methodology was validate from typical scenarios found in industrial environments, as star, line, cluster and mesh topologies. It was also evaluated scenarios with common cause failures and best practices to guide the design of an industrial wireless network. For guarantee scalability requirements, it was analyzed the performance of methodology in different scenarios where the results shown the applicability of proposal for networks typically found in industrial environments / Garantir os requisitos de dependabilidade ? fundamental para as aplica??es industriais, onde falhas podem conduzir a defeitos cujas consequ?ncias impactam em preju?zos econ?micos e principalmente danos ambientais e riscos aos operadores. Assim, diante da relev?ncia do tema, esta tese prop?em uma metodologia para an?lise da dependabilidade de redes industriais sem fio (WirelessHART, ISA100.11a, WIA-PA) ainda na fase de projeto. Entretanto, a proposta pode ser facilmente estendida para as fases de manuten??o e expans?o da rede. A proposta utiliza a teoria de grafos e o formalismo de ?rvores de Falhas para criar automaticamente um modelo anal?tico a partir de uma dada topologia de rede industrial sem fio, onde a dependabilidade possa ser avaliada. As m?tricas de avalia??o suportadas compreendem confiabilidade, disponibilidade e MTTF (tempo m?dio para falha) da rede, como tamb?m medidas de import?ncia dos dispositivos, aspectos de redund?ncia e defeitos em modo comum. Ressalta-se que a proposta independe de qualquer ferramenta para analisar quantitativamente as m?tricas visadas. Contudo, para prop?sito de valida??o da proposta utilizou-se uma ferramenta amplamente aceita na academia para esse fim (SHARPE). Adicionalmente, um algoritmo para gera??o dos cortes m?nimos originalmente aplicado na teoria de grafos foi adaptado para o formalismo das ?rvores de Falhas com o objetivo de garantir escalabilidade da metodologia ?s redes industriais sem fio (< 100 dispositivos). Finalmente, a metodologia proposta foi validada a partir de cen?rios t?picos encontrados em ambientes industriais, como topologias estrela, linha, cluster e mesh. Foram tamb?m avaliados cen?rios com defeitos em modo comum e um conjunto de pol?ticas a serem seguidas na cria??o de uma rede industrial sem fio. Para garantir aspectos de escalabilidade, uma an?lise de desempenho foi conduzida, onde pode-se observar a aplicabilidade da metodologia para as redes tipicamente encontradas em ambientes industriais