Pour répondre aux besoins de précision et d'efficacité des simulations scientifiques, la communauté du Calcul Haute Performance augmente progressivement les demandes en terme de parallélisme, rajoutant ainsi un besoin croissant de réutiliser les bibliothèques parallèles optimisées pour les architectures complexes.L'utilisation simultanée de plusieurs bibliothèques de calcul parallèle au sein d'une application soulève bien souvent des problèmes d 'efficacité. En compétition pour l'obtention des ressources, les routines parallèles, pourtant optimisées, se gênent et l'on voit alors apparaître des phénomènes de surcharge, de contention ou de défaut de cache.Dans cette thèse, nous présentons une technique de cloisonnement de flux de calculs qui permet de limiter les effets de telles interférences. Le cloisonnement est réalisé à l'aide de contextes d'exécution qui partitionnement les unités de calculs voire en partagent certaines. La répartition des ressources entre les contextes peut être modifiée dynamiquement afin d'optimiser le rendement de la machine. A cette fin, nous proposons l'utilisation de certaines métriques par un superviseur pour redistribuer automatiquement les ressources aux contextes. Nous décrivons l'intégration des contextes d'ordonnancement au support d'exécution pour machines hétérogènes StarPU et présentons des résultats d'expériences démontrant la pertinence de notre approche. Dans ce but, nous avons implémenté une extension du solveur direct creux qr mumps dans la quelle nous avons fait appel à ces mécanismes d'allocation de ressources. A travers les contextes d'ordonnancement nous décrivons une nouvelle méthode de décomposition du problème basée sur un algorithme de \proportional mapping". Le superviseur permet de réadapter dynamiquement et automatiquement l'allocation des ressources au parallèlisme irrégulier de l'application. L'utilisation des contextes d'ordonnancement et du superviseur a amélioré la localité et la performance globale du solveur. / To face the ever demanding requirements in term of accuracy and speed of scientific simulations, the High Performance community is constantly increasing the demands in term of parallelism, adding thus tremendous value to parallel libraries strongly optimized for highly complex architectures.Enabling HPC applications to perform efficiently when invoking multiple parallel libraries simultaneously is a great challenge. Even if a uniform runtime system is used underneath, scheduling tasks or threads coming from dfferent libraries over the same set of hardware resources introduces many issues, such as resource oversubscription, undesirable cache ushes or memory bus contention.In this thesis, we present an extension of StarPU, a runtime system specifically designed for heterogeneous architectures, that allows multiple parallel codes to run concurrently with minimal interference. Such parallel codes run within scheduling contexts that provide confined executionenvironments which can be used to partition computing resources. Scheduling contexts can be dynamically resized to optimize the allocation of computing resources among concurrently running libraries. We introduced a hypervisor that automatically expands or shrinks contexts using feedback from the runtime system (e.g. resource utilization). We demonstrated the relevance of this approach by extending an existing generic sparse direct solver (qr mumps) to use these mechanisms and introduced a new decomposition method based on proportional mapping that is used to build the scheduling contexts. In order to cope with the very irregular behavior of the application, the hypervisor manages dynamically the allocation of resources. By means of the scheduling contexts and the hypervisor we improved the locality and thus the overall performance of the solver.

The requirement of high quality services is increasing day by day. So, in order to meet up with this requirement new technologies are being developed one of them being virtualization. The main agenda of introducing virtualization is that though virtualization needs more powerful devices to run the hypervisor, the technique also helps to increase consolidation which makes efficient use of resources like increase in the CPU utilization. The virtualization technique helps us to run more VM’s (Virtual Machine) on the same platform i.e. on the same hypervisor. In virtualization as number of VM’s share the CPU will there be any effect on the performance of TCP with the performance influencing factors of virtualization. While TCP being the most widely used protocol and most reliable protocol can performance of TCP vary if different TCP congestion control mechanism are used in the virtualized environment are the main aims of this research.   In this study, we investigate the performance influencing factor of TCP in the virtualized environment and whether those influencing factors have any role to play with the performance of the TCP. Also which TCP congestion control mechanism is best suitable in order to download files when virtualization is used will be investigated by setting up a client-server test bed. The different TCP congestion control mechanism which have been used are CUBIC, BIC, Highspeed, Vegas, Veno, Yeah, Westwood, LP, Scalable, Reno, Hybla. Total download time has been compared in order to know which congestion control algorithm performs better in the virtualized environment.   The method that has been used to carry out the research is by experimentation. That is by changing the RAM sizes and CPU cores which are the performance influencing factors in virtualization and then analyzing the total download time while downloading a file by changing the TCP congestion control mechanisms by running a single guest VM. Apart from changing only congestion control mechanisms the other network parameters which effect the performance of the TCP such as Delay have been injected while downloading the file, to match up with the real time scenarios.   Results collected include average download time of a file by changing the different memory sizes and different CPU cores. Average Download time for different TCP congestion controls mechanisms with inclusion of the parameter that effects the total download time such as Delay.   From the results we got we can see that there is a slight influence on the performance of TCP by the performance influencing factors memory sizes and CPU cores allotted to the VM in the KVM virtualized environment and of all the TCP congestion control algorithms having TCP – BIC and TCP- YEAH performs the best in the KVM virtualized environment. The performance of TCP – LP is the least in the KVM virtualized environment.

Unikernels are lightweight, single-purpose virtual machines designed for the cloud. They provide enhanced security, minimal resource utilisation, fast boot times, and the ability to optimize performance for the target application. Despite their numerous advantages, unikernels face significant barriers to their widespread adoption. We identify two such obstacles as unscalable boot procedures in hypervisors and the difficulty in porting native applications to unikernel models. This work presents a solution for the first based on the popular Xen hypervisor, and demonstrates a significant performance benefit when running a large number of guest VMs. The HermiTux unikernel aims to overcome the second obstacle by providing the ability to run unmodified binaries as unikernels. This work adds to HermiTux, enabling it to retain some of the important advantages of unikernels such as fast system calls and modularity. / MS / Cloud computing provides economic benefits to users by allowing them to pay only for the resources that they use. Traditional virtual machines, so far the mainstay of cloud computing, come with a large number of features that are unnecessary for most cloud applications. Unikernels are specialised virtual machines that are compiled with only the features required to run the target application on top of a hypervisor. They have reduced memory requirements, short boot times, fast system calls, enhanced security and greater customizability. Despite these advantages, unikernels have not gained significant traction in industry. One reason is that existing hypervisors were not designed with unikernels in mind. Specifically, we show that for the Xen hypervisor, boot times rise exponentially with the number of VMs running on the system. The small size of unikernels allows us to run a much larger number of guest VMs than was previously possible, but these rising boot times present a major bottleneck. This thesis analyses the cause of this overhead and presents a solution that leads to a 4x reduction in the overall time required to boot 500 unikernels at once. Another reason for the slow adoption of unikernels is the difficulty involved in porting legacy applications to unikernel models. The HermiTux unikernel aims to remove this effort by allowing users to run unmodified, statically compiled executables compiled for Linux. In doing so, however, we lose the ability to modularise the unikernel for the application concerned, and also reintroduce a major source of overhead from regular applications - namely system calls. This thesis presents techniques based on binary analysis and binary rewriting that enable us to regain these advantages of unikernels in HermiTux.

Shared memory systems, such as SMP and ccNUMA topologies, simplify programming and administration. On the other hand, systems without hardware support for shared memory, such as clusters of commodity workstations, are commonly used due to cost and flexibility considerations. In this thesis, virtualisation is proposed as a technique that can bridge the gap between these architectures. The resulting system, vNUMA, is a hypervisor with a unique feature: it provides the illusion of shared memory across separate nodes on a fast network. This allows a cluster of workstations to be transformed into a single shared memory multiprocessor, supporting existing operating systems and applications. Such an approach could also have applications for emerging highly-parallel architectures, allowing a shared memory programming model to be retained while reducing hardware complexity. To build such a system, it is necessary to meld both a high-performance hypervisor and a high-performance distributed shared memory (DSM) system. This thesis addresses the challenges inherent in both of these tasks. First, designing an efficient hypervisor layer is considered; since vNUMA is implemented on the Itanium processor architecture, this is with particular reference to Itanium processor virtualisation. Then, novel DSM protocols are developed that allow SMP consistency models to be reproduced while providing better performance than a simple atomically-consistent DSM system. Finally, the system is evaluated, proving that it can provide good performance and compelling advantages for a variety of applications.

Shared memory systems, such as SMP and ccNUMA topologies, simplify programming and administration. On the other hand, systems without hardware support for shared memory, such as clusters of commodity workstations, are commonly used due to cost and flexibility considerations. In this thesis, virtualisation is proposed as a technique that can bridge the gap between these architectures. The resulting system, vNUMA, is a hypervisor with a unique feature: it provides the illusion of shared memory across separate nodes on a fast network. This allows a cluster of workstations to be transformed into a single shared memory multiprocessor, supporting existing operating systems and applications. Such an approach could also have applications for emerging highly-parallel architectures, allowing a shared memory programming model to be retained while reducing hardware complexity. To build such a system, it is necessary to meld both a high-performance hypervisor and a high-performance distributed shared memory (DSM) system. This thesis addresses the challenges inherent in both of these tasks. First, designing an efficient hypervisor layer is considered; since vNUMA is implemented on the Itanium processor architecture, this is with particular reference to Itanium processor virtualisation. Then, novel DSM protocols are developed that allow SMP consistency models to be reproduced while providing better performance than a simple atomically-consistent DSM system. Finally, the system is evaluated, proving that it can provide good performance and compelling advantages for a variety of applications.

Traditional network stacks in monolithic kernels have reliability and security concerns. Any fault in a network stack affects the entire system owing to lack of isolation in the monolithic kernel. Moreover, the large code size of the network stack enlarges the attack surface of the system. A multiserver OS design solves this problem. In contrast to the traditional network stack, a multiserver OS pushes the network stack into the network server as a user process, which performs three enhancements: (i) allows the network server to run in user mode while having its own address space and isolating any fault occurring in the network server; (ii) minimizes the attack surface of the system because the trusted computing base contracts; (iii) enables failure recovery, which is an important feature supported by a multiserver OS. This thesis proposes a network server for LibrettOS, an operating system based on rumprun unikernels and the Xen Hypervisor developed by Virginia Tech. The proposed network server is a service domain providing an L2 frame forwarding service for application domains and based on rumprun such that the existing device drivers of NetBSD can be leveraged with little modification. In this model, the TCP/IP stack runs directly in the address space of applications. This allows retaining the client state even if the network server crashes and makes it possible to recover from a network server failure. We leverage the Xen PCI passthrough to access a NIC (Network Interface Controller) from the network server. Our experimental evaluation demonstrates that the performance of the network server is good and comparable with Linux and NetBSD. We also demonstrate the successful recovery after a failure. / This research is based upon work supported by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA). The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the ODNI, IARPA, or the U.S. Government. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon. This research is also based upon work supported by the Office of Naval Research (ONR) under grants N00014-16-1-2104, N00014-16-1-2711, and N00014-18-1-2022. / Master of Science / When it comes to reliability and security in networking systems, concerns have been shown in traditional operating systems (OSs) such as Windows, MacOS, NetBSD, and Linux. Any fault in a networking system can have impacts on the entire system owing to lack of isolation in the OSs. Moreover, the large code size of a networking system enlarges the attack surface of the system. A multiserver OS design solves this problem by running a networking system as a network server, which performs three enhancements: (i) isolates any fault occurring in the network server itself; (ii) minimizes the attack surface of the system; and (iii) enables failure recovery. This thesis proposes a network server for LibrettOS, an operating system developed by Virginia Tech. The proposed network has two-pronged merits: (i) provides a system server providing a network packet forwarding service for applications; (ii) enables the existing device drivers of NetBSD to be leveraged with low amount of modification. Our experimental evaluation demonstrates that the performance of the network server outperforms state-of-the-art and comparable with Linux and that a successful recovery is possible after a failure.

Master’s thesis deals with the use of virtualization and consolidation technologies in order to optimize IT infrastructure in a selected company. The analysis contains current state of IT infrastructure and requirements for future upgrade. The theoretical part contains description of technologies and procedures used in virtualization and consolidation. Subsequently, the proposal of optimization and expansion of IT equipment is created together with management, implementation and economic evaluation of the solution.

Malware detection is typically performed using either software scanners running inside the operating system or external devices designed to validate the integrity of the kernel. This thesis proposes a hypervisor-based malware that compromises the system by targeting the hard disk drive and leaving the kernel unmodified. The hypervisor is able to issue read and write commands to the disk while actively hiding these actions from the operating system and any detection software therein. Additionally, the hypervisor’s presence has minimal impact on the performance of the system. The ability to perform these commands compromises the confidentiality, integrity, and availability of the stored data. As a result, this thesis has widespread implications affecting personal, corporate, and government users alike.

Denna avhandling utforskar den komplexa interaktionen mellan virtualiseringstekniker och prestanda för molnapplikationer. Genom att utvärdera hypervisor- och containerbaserade metoder skapar vi en teoretisk grund för att förstå rollen för virtualisering i modern molnbaserad miljö. Genom en metodisk förstudie och implementeringen av virtuella instanser samlar vi omfattande data om CPU-användning, nätverkstrafik, lagring och genomströmning, vilket avslöjar viktiga insikter. Resursanvändningen påverkas framför allt av hur databelastning ser ut, där resursintensiva uppgifter påverkar både hypervisor och containerinstanser. Studien visar att valet mellan hypervisor och containerteknik kan bero vissa användningsfall, vilket kräver ytterligare optimering. Att isolera tekniker på separata EC2- instanser ger bra kontroll men kräver löpande justeringar för varierande prestandamätningar. Slutsatserna hjälper till att förstå virtualiserings inverkan på molnapplikationer och betonar de avgörande rollerna för lagringshantering, nätverksdynamik och databelastning. Den här avhandlingen tar inte bara upp viktiga frågor, utan den lägger också grunden för framtida forskning som förutser förbättrad effektivitet och prestanda i virtualiserade molnmiljö. / This thesis explores the complex interaction between virtualization technologies and cloud application performance. By evaluating hypervisor and container-based approaches, we establish a theoretical foundation for understanding virtualization's role in modern cloud computing. Through a methodical pre-study and the implementation of virtual instances, we gather extensive data on CPU utilisation, network traffic, storage, and throughput to reveal key insights. Workloads notably impact resource usage, with resource-intensive tasks affecting both hypervisor and container instances. The study suggests that the choice between hypervisor and container technologies may depend on certain use cases, prompting further optimization. Isolating technologies on separate EC2 instances provides control but demands ongoing refinement for varied performance metrics. Conclusions contribute to understanding virtualization's impact on cloud applications, emphasising the critical roles of workload, network dynamics, and storage management. This thesis not only addresses key questions but also sets the stage for future explorations, promising enhanced performance and efficiency in virtualized cloud environments.

Industrial control systems are applied in many areas e.g., motion control for industrial robotics, process control of large plants such as in the area of oil and gas, and in large national power grids. Since the last decade with advancement and adoption of virtualization and multicore technology (e.g., Virtual Monitoring Machine, cloud computing, server virtualization, application virtualization), IT systems, automation industries have benefited from low investment, effective system management and high service availability. However, virtualization and multicore technologies have posed a serious challenge to real-time systems, which is violating timeliness and predictability of real-time application running on control systems. To address the challenge, we have extended a real-time component-based framework with virtual nodes; and evaluated the framework in the context of virtualized multicore environment. The evaluation is demonstrated by modeling and implementing an orchestra application with QoS for CPU, memory and network bandwidth. The orchestra application is a real-time and distributed application deployed on virtualized multicore PCs connected with speakers. The result shows undistorted orchestra performance played through speakers connected to physical computer nodes. The contribution of the thesis can be considered: 1) extending a real-time component-based framework, Future Automation Software Architecture (FASA) with virtual nodes using Virtual Computation Resource (VCR) and 2) design and installation of reusable test environment for development, debugging and testing of real-time application on a network of virtualized multicore environment. / Vinnova project “AUTOSAR for Multi-Core in Automotive and Automation Industries “