This thesis tackles the topic of serving IoT applications in the computing continuum. It proposes an approach to place applications in the tiers of the continuum, considering latency and energy as predefined metrics. It presents a system model to represent the computing continuum environment, and then, defines an optimization function that is tailored to meet the specific requirements of the IoT applications. The optimization function addresses the relationship between latency and energy consumption in the framework of IoT service provision, and it is implemented in two different directions: (1) the first direction uses a modified Genetic algorithm, and (2) the second direction utilizes the Machine learning concept. To evaluate the performance of the proposed approach, we incorporate different testbed setups and network configurations. All the setups and configurations are designed to represent the diverse demands of IoT applications. Then, different algorithms (such as Non-dominated Sorting Genetic Algorithm (NSGA), Brute Force, and Machine Learning) are implemented to provide different application placement scenarios. The results highlight the efficiency of the proposed approach in comparison with the Brute Force optimal solution while meeting the application requirements. This thesis proposes an optimized solution for serving IoT applications in the computing continuum environment. It considers two essential metrics (latency and energy consumption) in the applications placement processes while meeting the diverse functional and non-functional requirements of these applications. The study provides insights and ideas for future research to refine strategies that will minimize latency and energy consumption. It also urges researchers to consider more metrics while developing and implementing IoT applications. The requirements related to computing resources and performance levels make the development and implementation of these applications complex and challenging. This study serves as a foundational stepping stone towards addressing those challenges.

<p dir="ltr">The onset of smart manufacturing signifies a crucial shift in the industrial landscape, underscoring the pressing need for systems capable of adapting to and managing the complex dynamics of modern production environments. In this context, the importance of smart monitoring becomes increasingly apparent, serving as a vital tool for ensuring operational efficiency and reliability. Inspired by the critical role of auditory perception in human decision-making, this study investigated the application of machine sound recognition for practical use in manufacturing environments. Addressing the challenge of utilizing machine sounds in the loud noises of factories, the study employed an Internal Sound Sensor (ISS).</p><p dir="ltr">The study examined how sound propagates through structures and further explored acoustic characteristics of the ISS, aiming to apply these findings in machine monitoring. To leverage the ISS effectively and achieve a higher level of monitoring, a smart sound monitoring framework was proposed to integrate sound monitoring with machine data and human-machine interface. Designed for applicability and cost effectiveness, this system employs real-time edge computing, making it adaptable for use in various industrial settings.</p><p dir="ltr">The proposed framework and ISS deployed across a diverse range of production environments, showcasing a leap forward in the integration of smart technologies in manufacturing. Their application extends beyond continuous manufacturing to include discrete manufacturing systems, demonstrating adaptability. By analyzing sound signals from various production equipment, this study delves into developing machine sound recognition models that predict operational states and productivity, aiming to enhance manufacturing efficiency and oversight on real factory floors. This comprehensive and practical approach underlines the framework's potential to revolutionize operational management and manufacturing productivity. The study progressed to integrating manufacturing context with sound data, advancing towards high-level monitoring for diagnostic predictions and digital twin. This approach confirmed sound recognition's role in manufacturing diagnostics, laying a foundation for future smart monitoring improvements.</p>

Deep Learning and Neural Networks have been studied and developed for many years as of today, but there is still a great need of research on this field, because the industry needs are rapidly changing. The new challenge in this field is called edge inference and it is the deployment of Deep Learning on small, simple and cheap devices, such as low-power microcontrollers. At the same time, also on the field of hardware design the industry is moving towards the RISC-V micro-architecture, which is open-source and is developing at such a fast rate that it will soon become the standard. A batteryless ultra low power microcontroller based on energy harvesting and RISC-V microarchitecture has been the final target device of this thesis. The challenge on which this project is based is to make a simple Neural Network work on this chip, i.e., finding out the capabilities and the limits of this chip for such an application and trying to optimize as much as possible the power and energy consumption. To do that TensorFlow Lite Micro has been chosen as the Deep Learning framework of reference, and a simple existing application was studied and tested first on the SparkFun Edge board and then successfully ported to the RISC-V ONiO.zero core, with its restrictive features. The optimizations have been done only on the convolutional layer of the neural network, both by Software, implementing the Im2col algorithm, and by Hardware, designing and implementing a new RISC-V instruction and the corresponding Hardware unit that performs four 8-bit parallel multiply-and-accumulate operations. This new design drastically reduces both the inference time (3.7 times reduction) and the number of instructions executed (4.8 times reduction), meaning lower overall power consumption. This kind of application on this type of chip can open the doors to a whole new market, giving the possibility to have thousands small, cheap and self-sufficient chips deploying Deep Learning applications to solve simple everyday life problems, even without network connection and without any privacy issue. / Deep Learning och neurala nätverk har studerats och utvecklats i många år fram till idag, men det finns fortfarande ett stort behov av forskning på detta område, eftersom industrins behov förändras snabbt. Den nya utmaningen inom detta område kallas edge inferens och det är implementeringen av Deep Learning på små, enkla och billiga enheter, såsom lågeffektmikrokontroller. Samtidigt, även på området hårdvarudesign, går industrin mot RISC-V-mikroarkitekturen, som är öppen källkod och utvecklas i så snabb takt att den snart kommer att bli standarden. En batterilös mikrokontroller med ultralåg effekt baserad på energiinsamling och RISC-V-mikroarkitektur har varit den slutliga målenheten för denna avhandling. Utmaningen som detta projekt är baserat på är att få ett enkelt neuralt nätverk att fungera på detta chip, det vill säga att ta reda på funktionerna och gränserna för detta chip för en sådan applikation och försöka optimera så mycket som möjligt ström- och energiförbrukningen. För att göra det har TensorFlow Lite Micro valts som referensram för Deep Learning, och en enkel befintlig applikation studerades och testades först på SparkFun Edge-kortet och portades sedan framgångsrikt till RISC-V ONiO.zero-kärnan, med dess restriktiva funktioner. Optimeringarna har endast gjorts på det konvolutionerande skikt av det neurala nätverket, både av mjukvara, implementering av Im2col-algoritmen, och av hårdvara, design och implementering av en ny RISC-V-instruktion och motsvarande hårdvaruenhet som utför fyra 8-bitars parallella multiplikation -och-ackumulationsoperationer. Denna nya design minskar drastiskt både slutledningstiden (3,7 gånger kortare) och antalet utförda instruktioner (4.8 gånger färre), vilket innebär lägre total strömförbrukning. Den här typen av applikationer på den här typen av chip kan öppna dörrarna till en helt ny marknad, vilket ger möjlighet att ha tusentals små, billiga och självförsörjande chip som distribuerar Deep Learning-applikationer för att lösa enkla vardagsproblem, även utan nätverksanslutning och utan någon integritetsproblematik.

Immersion based technologies such as Augmented Reality (AR), Virtual Reality (VR) and Mixed Reality (MR), together falling under the umbrella of Extended Reality (XR) have taken the world by storm in the recent past. However, with the growing market and the increasing number of applications of XR, multiple challenges have arisen. To maintain acceptable levels of motion-to-photon latency, there is a need to serve the users with ultra low latency and with high reliability. To provide high quality rendering, these solutions have traditionally been deployed with wired connections, but severely inhibiting user mobility. Thus, the need to develop wireless solutions promising ultra low latency and high reliability emerges. Cloud/Edge based solutions promise to provide great dividends in this regard but it still remains crucial to understand how different scheduling policies perform against one another in terms of average throughput, mean system time, the number of UEs which can be serviced simultaneously etc. In this thesis, we explore how online packet scheduling policies such as first-come-first-serve, earliestdeadline-first, maximum weight scheduling etc. compare against other Quality of Experience(QoE)/ packet weight aware online scheduling policies and also against optimal offline schemes such as maximum-weighted-bipartitematching. We perform a detailed analysis of how these policies fare by studying various metrics such as the average-packet system time, competitive ratios, packet drop percentages and weight throughput, amongst others. Finally, we also explore how the introduction of multi-layered video encoding impacts XR service. Amongst the findings of the thesis, we conclude that it is possible to come up with solutions such as EDFα (which is a deadline and weight aware derivative of the earliest deadline first scheduling policy), which can either increase the weight throughput when compared to other baselines while also providing lesser packet drops and lower average system times for the scheduled packets. This algorithm can be further tuned by varying α to accordingly alter the weight throughput, system time and packet drop ratio depending on the precise user application. Additionally, we establish with the help of simulations that the introduction of multi-layered video encoding conclusively helps in reducing the average system time and eventually allows for more users to be accommodated in an XR based system at the cost of worsening video quality. / Immersionsbaserade teknologier som Augmented Reality (AR), Virtual Reality (VR) och Mixed Reality (MR), som tillsammans faller under paraplyet Extended Reality (XR) har tagit världen med storm på senare tid. Men med den växande marknaden och det ökande antalet tillämpningar av XR har flera utmaningar uppstått. För att förhindra åksjuka hos användare och för att upprätthålla acceptabla nivåer av rörelse-till-foton-latens, finns det ett behov av att betjäna användarna med ultralåg latens och med hög tillförlitlighet. För att ge högkvalitativ rendering har dessa lösningar traditionellt implementerats med trådbundna anslutningar, men de hämmar kraftigt användarens rörlighet. Därför uppstår behovet av att utveckla trådlösa lösningar som lovar ultralåg latens och hög tillförlitlighet. Moln/Edge-baserade lösningar lovar att ge stor utdelning i detta avseende, men det är fortfarande viktigt att förstå hur olika schemaläggningspolicyer fungerar mot varandra när det gäller genomsnittlig genomströmning, genomsnittlig systemtid, antalet UE:er som kan betjänas samtidigt etc. I den här avhandlingen undersöker vi hur online-paketschemaläggningspolicyer som round robin, först till kvarnförst-kvarn, tidigast-deadline-först, schemaläggning för maximal vikt etc. jämförs med andra Quality of Experience (QoE)/Viktmedvetna onlineschemaläggningspolicyer och även mot optimala offline-scheman såsom maximalt viktad-bipartite-matchning. Vi utför en detaljerad analys av hur dessa policyer klarar sig genom att studera olika mätvärden, såsom den genomsnittliga paketets systemtid, konkurrensförhållanden, procentsatser för paketnedgång och viktad genomströmning, bland annat. Slutligen undersöker vi också hur introduktionen av flerskiktad videokodning påverkar XRtjänsten. Bland resultaten av avhandlingen drar vi slutsatsen att det är möjligt att komma med lösningar som EDFα (som är en deadline- och viktmedveten derivata av Earliest deadline first scheduling policy), som antingen kan öka den viktade genomströmning jämfört med andra baslinjer samtidigt som det ger mindre paketnedgångar och lägre genomsnittliga systemtider för de schemalagda paketen. Denna algoritm kan ställas in ytterligare genom att variera α för att följaktligen ändra den viktade genomströmningen, systemtiden och paketnedgångshastigheten beroende på den exakta användarapplikationen. Dessutom fastställer vi med hjälp av simuleringar att införandet av flerskiktsvideokodning definitivt hjälper till att minska den genomsnittliga systemtiden och så småningom tillåter fler användare att få plats i ett XR-baserat system.

Abstract Communication networks are experiencing a significant growth of data traffic posing new challenges to the overall systems that should become more reactive and adaptive towards dynamically changing traffic, connections and network conditions. This thesis examines adaptive traffic management solutions within heterogeneous communication networks, which can be utilized to improve network performance, provide Quality of Service (QoS) for traffic paths and share resources in a fair way. The developed adaptive methods include solutions for fuzzy flow scheduling (AWFQ, FWQ) and regressive admission control (REAC) to provide stable network performance and efficient resource control. Such techniques for adaptive traffic management continuously balance and control traffic usage and recover from network faults and attacks. The results utilize traffic monitoring for estimating the overall network conditions, applying cognition to learn from previous actions, and adapting to the current traffic conditions for resource optimization. The thesis researches how to distribute these computing mechanisms towards network edges closer to the actual application users for more efficient resource usage, and to provide better performance for delay-sensitive applications. The methods developed have been applied to vehicular communications to assess and improve the messaging between vehicles and vulnerable road users (VRUs). These mechanisms are able to react faster to data traffic changes and guarantee better quality for prioritized traffic and users while at the same time they preserve fairness to other flows compared to traditional control and scheduling methods without adaptive characteristics. The overall system reacts to changes in the network QoS by determining decision-making procedures on possible flow rejection, marking, or allowed bandwidth weight assignment, thus bringing cognition to the network path. / Tiivistelmä Merkittävä liikennemäärien kasvu aiheuttaa tietoverkoille uusia haasteita, minkä vuoksi niiden täytyy tukea reaktiivisuutta ja adaptiivisuutta vastatakseen muuttuviin liikenne- sekä verkko-olosuhteisiin että yhteyksiin. Väitöskirjassa kehitetään heterogeenisten tietoverkkojen adaptiivisia liikenteenhallintaratkaisuja, joita voidaan hyödyntää verkon suorituskyvyn parantamiseen, tarjoamaan liikenteen palvelunlaatua (QoS) sekä tasapuolista resurssien jakoa. Kehitetyt adaptiiviset menetelmät sisältävät ratkaisuja sumeaan logiikkaan perustuvaan skedulointiin sekä regressiiviseen verkon pääsynhallintaan pohjautuen, jotka takaavat vakaamman verkon suorituskyvyn ja resurssien hallinnan. Nämä menetelmät tasapainottavat ja kontrolloivat liikennettä sekä pyrkivät palautumaan verkon häiriöistä ja hyökkäyksistä. Tulokset hyödyntävät liikenteen monitorointia verkon tilan arviointiin, soveltavat kognitiivisuutta oppiakseen aiemmista toiminnoista sekä adaptoituvat nykytilanteeseen resurssien optimoimiseksi. Väitöskirja tutkii, miten kyseisiä laskentamenetelmiä voidaan hajauttaa verkon reunoille lähemmäksi sovellusten käyttäjiä resurssien käytön tehostamiseksi sekä tarjoamaan parempaa suorituskykyä viiveherkille sovelluksille. Kehitettyjä menetelmiä sovelletaan autoverkkoihin autojen sekä suojattomien tienkäyttäjien viestinnän määrittämiseen sekä parantamiseen. Nämä menetelmät reagoivat nopeammin dataliikenteen muutoksiin, takaavat paremman laadun priorisoidulle liikenteelle sekä samalla tasapuolisuutta muulle liikenteelle verrattuna perinteisiin kontrollointi- ja skedulointimenetelmiin. Kehitetty järjestelmä reagoi verkon palvelunlaadun muutoksiin määrittelemällä päätöksentekomalleja mahdolliseen tietovuon hylkäämiseen, merkitsemiseen tai kaistankäytön painokertoimen määrittämiseen, täten luoden kognitiivisuutta verkon reitille.

Indiana University-Purdue University Indianapolis (IUPUI) / Research work presented within this thesis propose a neoteric variant of deep convolutional neural network architecture, CondenseNeXt, designed specifically for ARM-based embedded computing platforms with constrained computational resources. CondenseNeXt is an improved version of CondenseNet, the baseline architecture whose roots can be traced back to ResNet. CondeseNeXt replaces group convolutions in CondenseNet with depthwise separable convolutions and introduces group-wise pruning, a model compression technique, to prune (remove) redundant and insignificant elements that either are irrelevant or do not affect performance of the network upon disposition. Cardinality, a new dimension to the existing spatial dimensions, and class-balanced focal loss function, a weighting factor inversely proportional to the number of samples, has been incorporated in order to relieve the harsh effects of pruning, into the design of CondenseNeXt’s algorithm. Furthermore, extensive analyses of this novel CNN architecture was performed on three benchmarking image datasets: CIFAR-10, CIFAR-100 and ImageNet by deploying the trained weight on to an ARM-based embedded computing platform: NXP BlueBox 2.0, for real-time image classification. The outputs are observed in real-time in RTMaps Remote Studio’s console to verify the correctness of classes being predicted. CondenseNeXt achieves state-of-the-art image classification performance on three benchmark datasets including CIFAR-10 (4.79% top-1 error), CIFAR-100 (21.98% top-1 error) and ImageNet (7.91% single model, single crop top-5 error), and up to 59.98% reduction in forward FLOPs compared to CondenseNet. CondenseNeXt can also achieve a final trained model size of 2.9 MB, however at the cost of 2.26% in accuracy loss. Thus, performing image classification on ARM-Based computing platforms without requiring a CUDA enabled GPU support, with outstanding efficiency.

This thesis focuses on the Internet of Things and some of the most important problems it faces today. Among these are the overdependence on the Cloud and lack of autonomy, poor security and privacy, complicated initialization and power consumption. The work aims to implement a complex IoT solution that solves the discussed problems. The project is part of a collaboration with NXP Semicondutors and will be used to showcase the company's technologies.

<p>Online decision making with switching costs has received considerable attention in many practical problems that face uncertainty in the inputs and key problem parameters. Because of the switching costs that penalize the change of decisions, making good online decisions under such uncertainty is known to be extremely challenging. This thesis aims at providing new online algorithms with strong performance guarantees to address this challenge.</p> <p><br></p> <p>In part 1 and part 2 of this thesis, motivated by Network Functions Virtualization and smart grid, we study competitive online convex optimization with switching costs. Specifically, in part 1, we focus on the setting with an uncertainty set (one type of prediction) and hard infeasibility constraints. We develop new online algorithms that can attain optimized competitive ratios, while ensuring feasibility at all times. Moreover, we design a robustification procedure that helps these algorithms obtain good average-case performance simultaneously. In part 2, we focus on the setting with look-ahead (another type of prediction). We provide the first algorithm that attains a competitive ratio that not only decreases to 1 as the look-ahead window size increases, but also remains upper-bounded for any ratio between the switching-cost coefficient and service-cost coefficient.</p> <p><br></p> <p>In part 3 of this thesis, motivated by edge computing with artificial intelligence, we study bandit learning with switching costs where, in addition to bandit feedback, full feedback can be requested at a cost. We show that, when only 1 arm can be chosen at a time, adding costly full-feedback is not helpful in fundamentally reducing the Θ(<em>T</em>2/3) regret over a time-horizon <em>T</em>. In contrast, when 2 (or more) arms can be chosen at a time, we provide a new online learning algorithm that achieves a significantly smaller regret equal to <em>O</em>(√<em>T</em>), without even using full feedback. To the best of our knowledge, this type of sharp transition from choosing 1 arm to choosing 2 (or more) arms has never been reported in the literature.</p>

The industry is moving towards a higher degree of automation and connectivity, where previously manual operations are being adapted for interconnected industrial robots. This thesis focuses specifically on the automation of tightening applications with pre-tightened bolts and collaborative robots. The use of 3D computer vision is investigated for direct localization of bolts, to allow for flexible assembly solutions. A localization algorithm based on 3D data is developed with the intention to create a lightweight software to be run on edge devices. A restrictive use of deep learning classification is therefore included, to enable product flexibility while minimizing the computational load. The cloud-to-edge and cluster-to-edge trade-offs for the chosen application are investigated to identify smart offloading possibilities to cloud or cluster resources. To reduce operational delay, image partitioning to sub-image processing is also evaluated, to more quickly start the operation with a first coordinate and to enable processing in parallel with robot movement. Four different hardware architectures are tested, consisting of two different Single Board Computers (SBC), a cluster of SBCs and a high-end computer as an emulated local cloud solution. All systems but the cluster is seen to perform without operational delay for the application. The optimal hardware architecture is therefore found to be a consumer grade SBC, being optimized on energy efficiency, cost and size. If only the variance in communication time can be minimized, the cluster shows potential to reduce the total calculation time without causing an operational delay. Smart offloading to deep learning optimized cloud resources or a cluster of interconnected robot stations is found to enable increasing complexity and robustness of the algorithm. The SBC is also found to be able to switch between an edge and a cluster setup, to either optimize on the time to start the operation or the total calculation time. This offers a high flexibility in industrial settings, where product changes can be handled without the need for a change in visual processing hardware, further enabling its integration in factory devices. / Industrin rör sig mot en högre grad av automatisering och uppkoppling, där tidigare manuella operationer anpassas för sammankopplade industriella robotar. Denna masteruppsats fokuserar specifikt på automatiseringen av åtdragningsapplikationer med förmonterade bultar och kollaborativa robotar. Användningen av 3D-datorseende undersöks för direkt lokalisering av bultar, för att möjliggöra flexibla monteringslösningar. En lokaliseringsalgoritm baserad på 3Ddata utvecklas med intentionen att skapa en lätt mjukvara för att köras på Edge-enheter. En restriktiv användning av djupinlärningsklassificering är därmed inkluderad, för att möjliggöra produktflexibilitet tillsammans med en minimering av den behövda beräkningskraften. Avvägningarna mellan edge- och moln- eller klusterberäkning för den valda applikationen undersöks för att identifiera smarta avlastningsmöjligheter till moln- eller klusterresurser. För att minska operationell fördröjning utvärderas även bildpartitionering, för att snabbare kunna starta operationen med en första koordinat och möjliggöra beräkningar parallellt med robotrörelser. Fyra olika hårdvaruarkitekturer testas, bestående av två olika enkortsdatorer, ett kluster av enkortsdatorer och en marknadsledande dator som en efterliknad lokal molnlösning. Alla system utom klustret visar sig prestera utan operationell fördröjning för applikationen. Den optimala hårdvaruarkitekturen visar sig därmed vara en konsumentklassad enkortsdator, optimerad på energieffektivitet, kostnad och storlek. Om endast variansen i kommunikationstid kan minskas visar klustret potential för att kunna reducera den totala beräkningstiden utan att skapa operationell fördröjning. Smart avlastning till djupinlärningsoptimerade molnresurser eller kluster av sammankopplade robotstationer visar sig möjliggöra ökad komplexitet och tillförlitlighet av algoritmen. Enkortsdatorn visar sig även kunna växla mellan en edge- och en klusterkonfiguration, för att antingen optimera för tiden att starta operationen eller för den totala beräkningstiden. Detta medför en hög flexibilitet i industriella sammanhang, där produktändringar kan hanteras utan behovet av hårdvaruförändringar för visuella beräkningar, vilket ytterligare möjliggör dess integrering i fabriksenheter.

Serverless edge computing is a distributed network and computing system in which the data is processed at the edge of the network based on serverless architecture. It can provide large-scale computing and storage resources with low latency, which are very useful in AI applications such as object detection. However, when analyzing serverless computing architectures, we model them using simple models, such as single server or multi-server queues, and it is important to make sure these models can explain the behaviors of real systems. Therefore, we focus on the performance evaluation of serverless edge computing for AI applications in this project. With that, we aim at proposing more realistic and accurate models for real serverless architectures. In this project, our objective is to evaluate the performance and model mathematically an object-detection application running on a serverless architecture implemented with Kubernetes. This project provides a detailed description of the implementation of the serverless platform and YOLOv5-based object detection application. After implementation, we design experiments and make performance evaluations of the time of object detection results and quality of object detection results. Finally, we conclude that the number of users in the system significantly affects the service time. We observe that there is no queue in the system, so we cannot just use mathematical models with a queue to model the system. Therefore, we consider that the processor sharing model is more appropriate for modeling this serverless architecture. This is very helpful for giving insights on how to make more realistic and accurate mathematical queueing models for serverless architectures. For future work, other researchers can also implement our serverless platform and do further development, such as deploying other serverless applications on it and making performance evaluations. They can also design other use-cases for the experiments and make further analyses on queue modeling of serverless architecture based on this project. / Serverless edge computing är ett distribuerat nätverk och datorsystem där data bearbetas i kanten av nätverket baserat på serverlös arkitektur. Det kan tillhandahålla storskaliga dator- och lagringsresurser med låg latens, vilket är mycket användbart i AI-applikationer som objektdetektering. Men när vi analyserar serverlösa datorarkitekturer modellerar vi dem med hjälp av enkla modeller, till exempel enstaka servrar eller köer med flera servrar, och det är viktigt att se till att dessa modeller kan förklara beteendet hos verkliga system. Därför fokuserar vi på prestandautvärdering av serverlös edge computing för AI-applikationer i detta projekt. Med det siktar vi på att föreslå mer realistiska och exakta modeller för riktiga serverlösa arkitekturer. I detta projekt är vårt mål att utvärdera prestandan och matematiskt modellera en objektdetekteringsapplikation som körs på en serverlös arkitektur implementerad med Kubernetes. Detta projekt ger en detaljerad beskrivning av implementeringen av den serverlösa plattformen och den YOLOv5-baserade objektdetekteringsapplikationen. Efter implementering designar vi experiment och gör prestandautvärderingar av tidpunkten för objektdetekteringsresultat och kvaliteten på objektdetekteringsresultaten. Slutligen drar vi slutsatsen att antalet användare i systemet avsevärt påverkar servicetiden. Vi observerar att det inte finns någon kö i systemet, så vi kan inte bara använda matematiska modeller med en kö för att modellera systemet. Därför anser vi att processordelningsmodellen är mer lämplig för att modellera denna serverlösa arkitektur. Detta är mycket användbart för att ge insikter om hur man gör mer realistiska och exakta matematiska kömodeller för serverlösa arkitekturer. För framtida arbete kan andra forskare också implementera vår serverlösa plattform och göra vidareutveckling, såsom att distribuera andra serverlösa applikationer på den och göra prestandautvärderingar. De kan även designa andra användningsfall för experimenten och göra ytterligare analyser av kömodellering av serverlös arkitektur utifrån detta projekt.