Soft multiprocessor systems exploit the plentiful computational resources available in field programmable devices. By virtue of their adaptability and ability to support coarse grained parallelism, they serve as excellent platforms for rapid prototyping and design space exploration of embedded multiprocessor applications. As complex applications emerge, careful mapping, processor and interconnect customization are critical to the overall performance of the multiprocessor system. In this thesis, we have developed an automated scalable framework to efficiently map applications written in a high-level programmer-friendly language to customizable soft-cores. The framework allows the user to specify the application in a high-level language called Streamit. After an initial analysis of the application, a soft multiprocessor system is generated automatically using a set of customizable SPREE processors which communicate with each other over point-to-point FIFO connections. Several micro-architectural features of the processors are then automatically customized on a per-application basis to improve system area, performance and power consumption. The efficiency and scalability of this approach has been validated using a diverse set of eight audio, video and signal processing benchmarks on soft multiprocessor systems consisting of one to sixteen processors. Results show that generated soft multiprocessor systems consisting of sixteen processors can offer up to 6x speedup over a conventional single processor system. Our experiments with soft multiprocessor interconnection networks show that point-to-point topologies perform approximately 2x better than mesh topologies. Finally, we demonstrate that application-specific customizations on the instruction set, memory size, and inter-processor buffer size can improve the area and performance of the generated soft multiprocessor systems. The developed framework facilitates rapid design space exploration of soft multiprocessors.

The IoT sensing system plays an important role in the field of the smart city. IoT devices are generally constrained nodes due to their limited power and memory. How to save energy has been a challenge for the scalability of sensing networks. Previous studies introduce the dynamic sink and three dynamic sink deployment strategies. It has been proved by simulation experiments that the sensing network with dynamic sinks can reduce energy consumption. Further investigations on new dynamic sink deployment strategies are needed to explore the full potential of dynamic sinks. This work investigates three new deployment strategies, namely Determinisitic Strategy, Prediction Strategy, and Improved Prediction Strategy. We design experiments with different scenarios and evaluate the packet delivery ratio (PDR) and power consumption performances using emulated IoT devices on the Cooja simulator. The results show that the setups with these three new deployment strategies have good performance in terms of PDR and power consumption. Furthermore, we compare the performance difference between these three new strategies. The Improved Prediction Strategy has advantages over the other two strategies and has application prospects in reality. / IoT-baserade sensorsystem spelar en viktig roll för smarta städer. IoT-enheter är i allmänhet begränsade noder vad gäller till exempel kraftförsörjning och minnesutrymme. Hur man kan spara energi har varit en utmaning för skalbarheten hos sensornätverk. I tidigare studier introduceras dynamiska sänknoder och tre strategier för utplacering av sådana sänknoder. Det har visat sig genom simuleringsexperiment att ett nätverk med dynamiska sänknoder kan minska energiförbrukningen. Ytterligare undersökningar av nya strategier för utplacering av sänknoder behövs för att utforska den fulla potentialen hos dynamiska sänknoder. I det här arbetet undersöks tre nya strategier, nämligen Determinisitic Strategy, Prediction Strategy och Improved Prediction Strategy. Vi utformar experiment med olika scenarier och utvärderar andelen levererade paket (Packet Delivery Ration", PDR) och energiförbrukningen med hjälp av emulerade IoT-enheter i Cooja-simulatorn. Resultaten visar att uppställningarna med dessa tre nya strategier har bra prestanda när det gäller PDR och energiförbrukning. Dessutom jämför vi prestandaskillnaden mellan dessa tre nya strategier. Improved Prediction Strategy har fördelar jämfört med de andra två strategierna och bedöms ha goda tillämpningsmöjligheter i verkliga miljöer.

The integration of The Internet of Things and cloud computing has led to the emergenceof new classes of applications ranging from smart healthcare, smart and precision agriculture,smart manufacturing to smart environmental monitoring. The rapid surge in the useof these applications is expected to generate massive amounts of data with differentcharacteristics that are yet not studied. It can be hypothesised that each IoT-enabledapplication may exhibit a diverse range of characteristics that if modelled correctly, maylead to efcient distributed systems. This thesis aims to study the trafc characteristics ofan IoT-enabled healthcare application to build intelligent policies for scalable IoT-cloudsystems by employing the use of workload prediction and load balancing demonstratedon CloudSim Plus platform. The realistic incoming trafc from the SSiO IoT healthcareapplication system is studied, developed and modeled. Workload prediction algorithmsare developed based on ARIMA and SARIMA. The workload prediction algorithms arethen performed and extensively evaluated to select the one with the best performance,which was SARIMA, outperforming ARIMA by 200% on the basis of MAE, RMSE andMAPE. On the basis of the SARIMA prediction for 2 time periods in advance, theload balancing algorithm is preempted to perform horizontal scaling. The results revealthat the load balancer with SARIMA prediction outperform round robin and active loadbalancers for response time and cost by atleast 64% when it comes to worst case scenario.To conclude, a reflection is commented upon about the load balancing for IoT systemsand the directions this could take in the future for a more holistic sustainable approachon real life platforms.

No description available.

Pattern matching on large graphs is the foundation for a variety of application domains. The continuously increasing size of the underlying graphs requires highly parallel in-memory graph processing engines that need to consider non-uniform memory access (NUMA) and concurrency issues to scale up on modern multiprocessor systems. To tackle these aspects, a fine-grained graph partitioning becomes increasingly important. Hence, we present a classification of graph partitioning strategies and evaluate representative algorithms on medium and large-scale NUMA systems in this paper. As a scalable pattern matching processing infrastructure, we leverage a data-oriented architecture that preserves data locality and minimizes concurrency-related bottlenecks on NUMA systems. Our in-depth evaluation reveals that the optimal partitioning strategy depends on a variety of factors and consequently, we derive a set of indicators for selecting the optimal partitioning strategy suitable for a given graph and workload.

This master's thesis investigates the application of Graph Neural Networks (GNNs) to address scalability challenges in combinatorial optimization, with a primary focus on the minimum Total Dominating set Problem (TDP) and additionally the related Carrier Scheduling Problem (CSP) in networks of Internet of Things. The research identifies the NP-hard nature of these problems as a fundamental challenge and addresses how to improve predictions on input graphs of sizes much larger than seen during training phase. Further, the thesis explores the instability in such scalability when leveraging GNNs for TDP and CSP. Two primary measures to counter this scalability problem are proposed and tested: incorporating node degree as an additional feature and modifying the attention mechanism in GNNs. Results indicate that these countermeasures show promise in addressing scalability issues in TDP, with node degree inclusion demonstrating overall performance improvements while the modified attention mechanism presents a nuanced outcome with some metrics improved at the cost of others. Application of these methods to CSP yields bleak results, evincing the challenges of scalability in more complex problem domains. The thesis contributes by detecting and addressing scalability challenges in combinatorial optimization using GNNs and provides insights for further research in refining methodologies for real-world applications. / Denna masteruppsats undersöker tillämpningen av Grafiska Neurala Nätverk (GNN) för att hantera utmaningar inom skalbarhet vid kombinatorisk optimering, med ett primärt fokus på minimum Total Dominating set Problem (TDP) samt även det relaterade Carrier Scheduling Problem (CSP) i nätverk inom Internet of Things. Studien identifierar den NP-svåra karaktären av dessa problem som en grundläggande utmaning och lyfter hur man kan förbättra prediktioner på indatagrafer av storlekar som är mycket större än vad man sett under träningsfasen. Vidare utforskar uppsatsen instabiliteten i sådan skalbarhet när man utnyttjar GNN för TDP och CSP. Två primära åtgärder mot detta skalbarhetsproblem föreslås och testas: inkorporering av nodgrad som ett extra attribut och modifiering av attention-mekanismer i GNN. Resultaten indikerar att dessa motåtgärder har potential för att angripa skalbarhetsproblem i TDP, där inkludering av nodgrad ger övergripande prestandaförbättringar medan den modifierade attention-mekanismen ger ett mer tvetydigt resultat med vissa mätvärden förbättrade på bekostnad av andra. Tillämpning av dessa metoder på CSP ger svaga resultat, vilket antyder om utmaningarna med skalbarhet i mer komplexa problemdomäner. Uppsatsen bidrar genom att upptäcka och adressera skalbarhetsutmaningar i kombinatorisk optimering med hjälp av GNN och ger insikter för vidare forskning i att förfina metoder för verkliga tillämpningar.

Last years, due to the advent of more sophisticated tools for exploratory data analysis, data management, Machine Learning (ML) model training and model serving into production, the concept of MLOps has gained more popularity. As an effort to bring DevOps processes to the ML lifecycle, MLOps aims at more automation in the execution of diverse and repetitive tasks along the cycle and at smoother interoperability between teams and tools involved. In this context, the main cloud providers have built their own ML platforms [4, 34, 61], offered as services in their cloud solutions. Moreover, multiple frameworks have emerged to solve concrete problems such as data testing, data labelling, distributed training or prediction interpretability, and new monitoring approaches have been proposed [32, 33, 65]. Among all the stages in the ML lifecycle, one of the most commonly overlooked although relevant is model monitoring. Recently, cloud providers have presented their own tools to use within their platforms [4, 61] while work is ongoing to integrate existent frameworks [72] into open-source model serving solutions [38]. Most of these frameworks are either built as an extension of an existent platform (i.e lack portability), follow a scheduled batch processing approach at a minimum rate of hours, or present limitations for certain outliers and drift algorithms due to the platform architecture design in which they are integrated. In this work, a scalable automated cloudnative architecture is designed and evaluated for ML model monitoring in a streaming approach. An experimentation conducted on a 7-node cluster with 250.000 requests at different concurrency rates shows maximum latencies of 5.9, 29.92 and 30.86 seconds after request time for 75% of distance-based outliers detection, windowed statistics and distribution-based data drift detection, respectively, using windows of 15 seconds length and 6 seconds of watermark delay. / Under de senaste åren har konceptet MLOps blivit alltmer populärt på grund av tillkomsten av mer sofistikerade verktyg för explorativ dataanalys, datahantering, modell-träning och model serving som tjänstgör i produktion. Som ett försök att föra DevOps processer till Machine Learning (ML)-livscykeln, siktar MLOps på mer automatisering i utförandet av mångfaldiga och repetitiva uppgifter längs cykeln samt på smidigare interoperabilitet mellan team och verktyg inblandade. I det här sammanhanget har de största molnleverantörerna byggt sina egna ML-plattformar [4, 34, 61], vilka erbjuds som tjänster i deras molnlösningar. Dessutom har flera ramar tagits fram för att lösa konkreta problem såsom datatestning, datamärkning, distribuerad träning eller tolkning av förutsägelse, och nya övervakningsmetoder har föreslagits [32, 33, 65]. Av alla stadier i ML-livscykeln förbises ofta modellövervakning trots att det är relevant. På senare tid har molnleverantörer presenterat sina egna verktyg att kunna användas inom sina plattformar [4, 61] medan arbetet pågår för att integrera befintliga ramverk [72] med lösningar för modellplatformer med öppen källkod [38]. De flesta av dessa ramverk är antingen byggda som ett tillägg till en befintlig plattform (dvs. saknar portabilitet), följer en schemalagd batchbearbetningsmetod med en lägsta hastighet av ett antal timmar, eller innebär begränsningar för vissa extremvärden och drivalgoritmer på grund av plattformsarkitekturens design där de är integrerade. I det här arbetet utformas och utvärderas en skalbar automatiserad molnbaserad arkitektur för MLmodellövervakning i en streaming-metod. Ett experiment som utförts på ett 7nodskluster med 250.000 förfrågningar vid olika samtidigheter visar maximala latenser på 5,9, 29,92 respektive 30,86 sekunder efter tid för förfrågningen för 75% av avståndsbaserad detektering av extremvärden, windowed statistics och distributionsbaserad datadriftdetektering, med hjälp av windows med 15 sekunders längd och 6 sekunders fördröjning av vattenstämpel.

No description available.

No description available.

Video conferencing has been a part of many communication platforms over the years. Over the last decades users have moved from dedicated telephony networks to the Internet, and recently to the Web. With the introduction of Web Real-Time Communication (WebRTC) it is now possible to make voice- and video calls simply by visiting a web page, without having to install any additional software. Services that enable multi-user conferences are quite common. However existing solutions such as the Multipoint Control Unit (MCU) inherently do not scale and can be a single point of failure, due to its centralized architecture. This can lead to high maintenance costs and poor service availability.To solve the scalability- and availability problems of video-conferencing services, a decentralized alternative to the MCU is proposed. A decentralized conferencing system uses the distributed resources of its users instead of relying on a central server. This means that the system can handle an increasing number of users without having to upgrade any server infrastructure. Additionally, failures are only partial and can happen regularly without affecting the rest of the system. This report presents the development of a protocol built on top of WebRTC that enables completely decentralized multi-user conferencing. It includes a distributed algorithm for voice-activated switching to reduce the computation and network resources used. A load-balancing technique based on media stream relays is used to distribute the resource requirements of the conference participants. The protocol is implemented as a Javascript library that can be included in a web application. A proof-of-concept web application is developed using the library and its performance is evaluated. The performance data is analyzed and the results are used to make incremental improvements to the protocol and implementation. Although not all features of the protocol are implemented, the tests show promising results. The application allows multiple users to participate in high-definition video conferences, with no server infrastructure aside from a Mini PC that hosts a web server and a WebRTC signaling server. / Videokonferenser har varit en del av många olika kommunikationsplattformar genom åren. Tekniken har yttats från dedikerade telefonnnät,, till Internet, och på senare tid till webben. I och med introduktionen av WebRTC (Web Real-Time Communication) är deti dag möjligt att enkelt deltaga i röst- och videosamtal genom att gå till en webbsida utan att behöva installera någon programvara annat an en webbläsare. De flesta existerande konferenstjänster är byggda med en centraliserad arkitektur, vilket kan leda till tekniska problem när antalet användare ökar eller när fel uppstår i systemens centrala servrar. Dessa problem kan leda till driftstopp och skada tjänstens tillgänglighet för användarna. Den här rapporten täcker utvecklingen av ett protokoll som tillsammans med WebRTC kan användas för att bygga en helt decentraliserad konferenstjänst. Målet är att tjänsten ska vara oberoende av centrala servrar, och på så vis lösa problemen med skalbarhet och tillgänglighet. Protokollet implementeras i en webbapplikation som testas och utvärderasöver flera iterationer för att hitta nya förbättringar. Testerna visar lovande resultat. Slutsatsen dras det är fullt möjligt att bygga en konferenstjänst på detta sätt, och möjligheter för framtida optimeringar och testfall föreslås.