The purpose of this dissertation describes several power optimization techniques for energy efficient datacenters. To achieve this goal, it approaches power dissipation holistically for entire datacenters and analyzes them layer-by-layer from (1) the infrastructure level, (2) the system level, and all the way down to (3) the micro-architecture level. First, for infrastructure-level power optimization of datacenters, this work presents infrastructure-level mathematical models and a holistic warehouse-scale datacenter power and performance simulator, SimWare. Experiments using SimWare show a high loss of cooling efficiency resulting from the non-uniform inlet air temperature distribution across servers. Second, this study describes a system-level technique, ATAC, which maximizes power efficiency while minimizing overheating. Finally, this dissertation describes a micro-architecture level technique under the context of emerging non-volatile memory technologies. We first show that storing more than one bit per cell, or multiple bits per cell, ends up with much higher soft-error rates than conventional technologies. However, multi-bit per cell technology can still be used as approximate storage. To this end, we propose a new class of multi-bit per cell memory in which both a precise bit and an approximate bit are located in a physical cell. With the development of these techniques, the contribution of this body of work is a reduction in the power consumption of datacenters in a holistic way, eliminating one of the most important hurdles to the proliferation of cloud-computing environments.

This thesis examines different methods for the cooling of data centers. Several means of cooling with the refrigeration cycle have been studied, as well as the feasibility of using free cooling in different parts of Sweden. In addition, the possibilities of transferring excess heat to a nearby district heating system were investigated. It was found that in all investigated cases except two, the data center could be cooled solely with free cooling for at least 300 days a year when only the outside air temperature was considered as a constraining factor. The limit was then set to 20 °C. The vast majority of all district heating systems in Sweden get little to none of their heat from other excess heat sources. It is therefore unlikely that this would hinder the possibility for owners of data centers to sell the heat produced by the servers to district heating systems. The data center belonging to the Swedish Energy Agency was especially examined. Experiments showed that when the cooling system temperature was set to 20 °C or below, the power consumption increased by 1 kW. No upper limit was found. Images taken with an IR-camera indicated that some mixing of hot and cold air takes place near the end of the aisle. Moreover, the difference in temperature between the upper and lower parts of the rack increased when the cooling system was set to higher temperatures, indicating further mixing of air.

The purpose of this thesis is to examine how investments are justified in the Swedish data center industry, with a focus on cooling equipment. This is achieved through semi-structured interviews with 20 different interviewees. These interviewees represent different facets of the industry and facilitate an understanding of which factors, such as: investment costs, operating costs, reliability, environmental impact and the influence of external and internal actors, come into play and to which degree they impact the investment decision-making process. The results show that primarily low operating cost in the form of low energy consumption is the driving factor when it comes to the procurement of new equipment. A lower environmental impact is a positive side-effect of lower energy consumption. But alone it does not justify an additional investment. Data center downtime has a negative impact financially and can also adversely affect various functions critical to society. Therefore, the dependability of the equipment is still a priority, but reliability is now considered to have reached such a level that it is considered acceptable and the focus of development has shifted towards energy efficiency. The interview study is complemented by an investment-theory review of articles and books on related subject matter. This subject matter is linked directly or indirectly to disposition of investment in the data center industry. The framework developed from the review is then compared to the interview results. Results indicate that decision and investment theory in many cases can also be applied to data center investments. Theories that are more environmentally and socially oriented, such as Triple Bottom Line, are not consistent with the empirical data from the interviews. The investment decisions are also influenced by the current high availability and high reliability of data centers, which is the reason why the industry is rife with inertia, conservatism and skepticism to new solutions. Consequently, the parallel between data centers and high risk organizations, HRO, brought a new approach to how priorities are evaluated within HROs, namely that it is possible to combine high reliability with a low environmental impact. The results of the study are used as the basis to establish a marketing plan for the company SEE Cooling, a startup company that supplies cooling equipment. This marketing plan is based on the models for: Network analysis, Porter's five forces analysis, SWOT analysis and Kotler's marketing mix. The resulting marketing plan offers proposals on how SEE Cooling shall navigate their business network to overcome conservatism and promote their products, and illustrates potential difficulties.

This thesis analyzes the performance of Quantized Congestion Notification (QCN) during data access from clustered servers in data centers. The reasons why QCN does not perform adequately in these situations are examined and several modifications are proposed to the protocol to improve its performance in these scenarios. The causes of QCN performance degradation are traced to flow rate variability, and it is shown that adaptive sampling at the switch and adaptive self-increase of flow rates at the QCN rate limiter significantly enhance QCN performance in a TCP Incast setup. The performance of QCN is compared against TCP modifications in a heterogeneous environment, and it is shown that modifications to QCN yield better performance. Finally, the performance of QCN with the proposed modifications is compared with that of unmodified QCN in other workloads to show that the modifications do not negatively affect QCN performance in general.

Data centers operations are notoriously energy-hungry, with the computing and cooling infrastructures drawing comparable amount of electrical power to operate. A direction to improve their efciency is to optimizethe cooling, in the sense of implementing cooling infrastructures controlschemes that avoid performing over-cooling of the servers.Towards this direction, this work investigates minimum cost linearquadratic control strategies for the problem of managing air cooled datacenters. We derive a physical and a black box model for a general datacenter, identify this model from real data, and then derive, present andtest in the eld a model based Linear-Quadratic Regulator (LQR) strategy that sets the optimal coolant temperature for each individual coolingunit. To validate the approach we compare the eld tests from the LQR strategy against classical Proportional-Integral-Derivative (PID) controlstrategies, and show through our experiments that it is possible to reducethe energy consumption with respect to the existing practices by severalpoints percent without harming the servers within the data center fromthermal perspectives.

This thesis examines the life cycles of a data centre and a local server to conclude which of them has the largest environmental impact. Data centres are warehouse-looking buildings with servers in them, hosting the so-called “cloud”. Storing your data and information in a data centre and using its software and other resources through the internet is called cloud computing. The more traditional alternative to cloud computing is to have your own servers locally. You can access these without the internet, but you must operate and service it by yourself. There are some benefits with using cloud computing instead of local servers. One of these is often said to be energy efficiency and by that less environmental impact. This thesis sets out to confirm or refute, that statement. To do that, a comparing Life Cycle Assessment (LCA) has been done. A comparing LCA is a method to compare two similar systems and how much environmental impact they both have had during their lifetime. To assess this the amount of carbon dioxide equivalents emissions per the number of virtual machines has been used as a functional unit. The assessment was made on a hypothetical data centre and a hypothetical local server. The results showed that the local server had around 9.5 more carbon dioxide emissions per virtual machine than the data centre which confirms that the data centre has the lesser environmental impact of the two. This was mostly because of a more efficient use of the servers and the fact that a data centre has the possibility to recover some of its waste heat.

The aim of this study is to determine whether Facebook's establishment in Luleå has increased welfare. The essay also gives possible suggestions on how the state can increase profitability when considering future establishments. The thesis focuses on the period when Facebook’s two data centers were established, that is 2011-2020. The thesis mainly uses data from various public inquiries and government agencies. An interview with a professor in electricity production is also conducted to create a deeper understanding of the server halls' impact on electricity production in Sweden.  To determine whether the establishment of Facebook has increased welfare, a cost-benefit analysis is used as a method and theory. Furthermore, data centers and their impacts will be defined in chapters three and four as a basis for the calculation. In the calculation, we will compare benefits and costs. Benefits need to exceed costs for the project to be profitable. Whether the establishment has been profitable is examined from three perspectives: overall, national, and regional. Unlike the national and regional calculations, only costs are considered in the overall calculation. Based on assumptions that Facebook's data center has a lifespan of 10 years, with a discount rate of 3.5 percent, it can be stated that the total cost for the project will be  - SEK 14 673 137 420 - H - W in the overall calculation. For Facebook's establishment to be profitable, the overall benefits need to exceed the costs. According to the national calculation, Facebook's establishment based on monetary posts. is profitable as the net present value is positive and equals SEK 12 636 712 128. The non-monetary effect, the risk of electricity shortage, leads to a decrease in the net present value in the national calculation. However, we believe that it is unlikely that the cost of the risk would exceed SEK 12 billion and thus we consider the calculation to remain profitable despite the risk. From a regional perspective, Facebook's establishment is profitable if the non-monetary benefits outweigh the costs of the destroyed landscape and the risk of electricity shortage. According to the sensitivity analysis, Facebook's establishment is a cost to the state as the costs for the public sector exceed the tax revenues. Marginal excess burden, MEB, is used to calculate whether the project loads the state budget. Since we choose to use a MEB of 0.3, it can be stated that the value of the overall calculation is increased with - SEK 2 855 377. The other effects in our calculations such as operating costs, maintenance costs, use of electricity, economic growth, and landscape image cause uncertainty in our calculations which is also described in the sensitivity analysis. Finally, the study proposes five different suggestions to have in mind when future data center establishments are being planned in Sweden.

En el presente trabajo se describe el análisis que se ha realizado para poder transformar la red LAN de un centro de datos tradicional, líder en servicios de Outsourcing en el Perú, en una infraestructura de red basada en software, el cual, va a ayudar a mejorar los tiempos de entrega de los servicios de red; así como, incrementar el índice de disponibilidad de ésta. La empresa en estudio, actualmente, tiene tres centros de datos ubicado en dos distritos de la ciudad de Lima, como la Molina y San Isidro, la gerencia en el Perú mediante la unidad de negocio Global Technology Services (GTS), responsable de los centros de datos en el país, está liderando este cambio tecnológico que traerá consigo una revolución en el mercado de centro de datos en el país. En este documento se encuentra el diseño de alto y bajo nivel de la red bajo la tecnología de software defined network (SDN) que se va a proponer, también los beneficios en tiempo que toma implementar una red LAN usando esta tecnología, adicionalmente, el equipamiento de red que se propone para este diseño; y, por último, el análisis económico que muestra la justificación de implementar esta nueva red bajo software defined network. Finalmente, se hacen las recomendaciones que se deben seguir si se desea implementar una red LAN basada en software, tanto para un centro de datos, como para un campus; así como, las ventajas si se toma la decisión de implementarla. / This paper describes the analysis that has been carried out to transform the LAN network of a traditional data center, leader in Outsourcing services in Peru, into a software-based network infrastructure, which, will help improve delivery times for network services; as well as increase its availability index. The company under study currently has three data centers located in two districts of the city of Lima, such as La Molina and San Isidro, the management in Peru through the Global Technology Services (GTS) business unit, responsible for the centers of data in the country, is leading this technological change that will bring about a revolution in the data center market in the country. This document contains the high and low level design of the network under the defined network software (SDN) technology that is going to be proposed, as well as the benefits in time it takes to implement a LAN using this technology, additionally, the equipment network proposed for this design; and, finally, the economic analysis that shows the justification for implementing this new network under defined network software. Finally, recommendations are made that should be followed if you want to implement a software-based LAN network, both for a data center and for a campus; as well as the advantages if the decision is made to implement it. / Tesis

Ce travail aborde la problématique des systèmes de refroidissement ou de contrôle thermique industriels. Nous avons particulièrement mis l'accent sur le refroidissement des serveurs informatiques. Une première partie consiste en l'étude des moyens d'amélioration des techniques de refroidissement existantes, tandis que la deuxième partie est une réflexion sur des techniques de refroidissement alternatives potentiellement plus efficaces et répondant aux demandes actuelles du contrôle thermique industriel. Dans le premier chapitre, nous analysons la bibliographie et la théorie relatives aux phénomènes physiques derrière les techniques de refroidissement étudiées. Ensuite, une classification des techniques de refroidissement est proposée en fin de chapitre. Ce chapitre a servi de base pour l'amélioration des technologies de refroidissement existantes et à la réflexion sur de nouvelles techniques plus efficaces. Le second chapitre porte sur l'optimisation d'une plaque froide, destinée au refroidissement des serveurs informatiques, en s'aidant d'un outil numérique et d'essais expérimentaux. Nous avons noté une augmentation des transferts de chaleur dans la plaque froide en utilisant des inserts, notamment ceux en forme de losange disposés en quinconce. A l'inverse, l'utilisation de certains nanofluides en tant que fluides caloporteurs ne semble pas apporter de gain significatif. Dans le troisième chapitre nous détaillons la démarche suivie pour la conception d'un dissipateur de chaleur basé sur une technologie caloducs, destiné au refroidissement des cartes électroniques. En premier lieu, nous présentons le modèle thermohydraulique de dimensionnement d'un caloduc cylindrique ; une étude paramétrique (géométrique, type de fluide,...) nous a permis d'identifier le jeu de paramètres donnant la meilleure performance du caloduc. En second lieu, nous évoquons les tests réalisés sur le dissipateur de chaleur à caloduc qui nous amènent à valider en partie le modèle thermohydraulique développé. Le dernier chapitre porte sur la réalisation et l'étude d'un démonstrateur pour le refroidissement des cartes électroniques par immersion dans un liquide à basse température de saturation. On commence par la mise en place et l'utilisation d'un modèle numérique pour la conception du démonstrateur, puis des tests expérimentaux sont réalisés. Les premiers résultats obtenus en utilisant le SES-36 comme fluide de travail sont assez prometteurs.Mots clés : modélisation, transfert de chaleur, refroidissement, datacenter, liquid-cooling, caloducs, échangeurs, nanofluides, ébullition en vase, simulation numérique / The present work is about industrial thermal control systems issues, a focus is done on IT servers cooling. The first part of this document is about cooling techniques optimization. The second part concerns investigations for new cooling techniques potentially more efficient and which can address today's challenges of industrial thermal control systems. In the first chapter, we have done a literature survey and a theoretical analysis of physical phenomena behind studied cooling techniques. Then, we have sorted those techniques according to some criteria. This chapter is a basis for optimization studies we have achieved in the second chapter and for our new cooling techniques investigations in the two last chapters. In the second chapter, we have presented a study about heat transfer enhancement in a cold plate developed for IT servers liquid-cooling. We have observed great heat transfer enhancement when adding lozenge shaped fins inside the cold plate channels. No gain was noted when using some nanofluids as heat transfer liquids. We have shown, in the third chapter, a conception algorithm of a heat spreading device destined for IT servers cooling. The heat spreader is based on a heatpipe technology. First of all, we have presented the thermohydraulic model for cylindrical heatpipes design. A parametric study (geometric, working fluids ...) showed the best combination to obtain the highest heatpipe performance. Then, the heat spreader has been tested and we have validated partially the heatpipe model. The last chapter of this work is a study of a demonstrator destined for cooling down electronic components by immersion in a dielectric fluid with a low saturation temperature. We first built a numerical model to design the demonstrator and then it has been tested. The first results are very encouraging, when using SES-36 as a working fluid.Keywords : modeling, heat transfer, cooling, datacenter, liquid-cooling, heat pipes, heat exchangers, nanofluids, pool boiling, numerical simulation

Un Data Center (DC) est un bâtiment dont le but est d'héberger des équipements informatiques pour fournir différents services Internet. Pour assurer un fonctionnement constant de ces équipements, le système électrique fournit de l'énergie, et pour les maintenir à une température constante, un système de refroidissement est nécessaire. Chacun de ces besoins doit être assuré en permanence, car la conséquence de la panne de l'un d'eux entraîne une indisponibilité de l'ensemble du système du DC, ce qui peut être fatal pour une entreprise.A notre connaissance, il n'existe pas de travaux d'étude sur l'analyse de sûreté de fonctionnement et de performance, prenant en compte l'ensemble du système du DC avec les différentes interactions entre ses sous-systèmes. Les études d'analyse existantes sont partielles et se concentrent sur un seul sous-système, parfois deux. L'objectif principal de cette thèse est de contribuer à l'analyse de sûreté de fonctionnement d'un Data Center. Pour cela, nous étudions, dans un premier temps, chaque sous-système (électrique, thermique et réseau) séparément, afin d'en définir ses caractéristiques. Chaque sous-système du DC est un système de production qui transforment les alimentations d'entrée (énergie pour le système électrique, flux d'air pour le système thermique, et paquets pour le réseau) en sorties, qui peuvent être des services Internet. Actuellement, les méthodes d'analyse de sûreté de fonctionnement existantes pour ce type de systèmes sont inadéquates, car l'analyse de sûreté doit tenir compte non seulement de l'état interne de chaque composant du système, mais également des différents flux de production qui circulent entre ces composants. Dans cette thèse, nous considérons une nouvelle technique de modélisation appelée Arbres de Production (AP) qui permet de modéliser la relation entre les composants d'un système avec une attention particulière aux flux circulants entre ces composants.La technique de modélisation en AP permet de traiter un seul type de flux à la fois. Son application sur le sous-système électrique est donc appropriée, car il n'y a qu'un seul type de flux (le courant électrique). Toutefois, lorsqu'il existe des dépendances entre les sous-systèmes, comme c'est le cas pour les sous-systèmes thermiques et les sous-systèmes de réseaux, différents types de flux doivent être pris en compte, ce qui rend l'application de la technique des APs inadéquate. Par conséquent, nous étendons cette technique pour traiter les dépendances entre les différents types de flux qui circulent dans le DC. En conséquence, il est facile d'évaluer les différents indicateurs de sûreté de fonctionnement du système global du DC, en tenant compte des interactions entre ses sous-systèmes. De plus, nous faisons quelques statistiques de performance. Nous validons les résultats de notre approche en les comparant à ceux obtenus par un outil de simulation que nous avons implémenté et qui est basé sur la théorie des files d'attente.Jusqu'à présent, les modèles d'arbres de production n'ont pas d'outils de résolution. C'est pourquoi nous proposons une méthode de résolution basée sur la Distribution de Probabilité de Capacité (Probability Distribution of Capacity - PDC) des flux circulants dans le système du DC. Nous implémentons également le modèle d'AP en utilisant le langage de modélisation AltaRica 3.0, et nous utilisons son simulateur stochastique dédié pour estimer les indices de fiabilité du système. Ceci est très important pour comparer et valider les résultats obtenus avec notre méthode d'évaluation. En parallèle, nous développons un outil qui implémente l'algorithme de résolution des APs avec une interface graphique basée qui permet de créer, éditer et analyser des modèles d'APs. L'outil permet également d'afficher les résultats et génère un code AltaRica, qui peut être analysé ultérieurement à l'aide du simulateur stochastique de l'outil AltaRica 3.0. / A Data Center (DC) is a building whose purpose is to host IT devices to provide different internet services. To ensure constant operation of these devices, energy is provided by the electrical system, and to keep them at a constant temperature, a cooling system is necessary. Each of these needs must be ensured continuously, because the consequence of breakdown of one of them leads to an unavailability of the whole DC system, and this can be fatal for a company.In our Knowledge, there exists no safety and performance studies’, taking into account the whole DC system with the different interactions between its sub-systems. The existing analysis studies are partial and focus only on one sub-system, sometimes two. The main objective of this thesis is to contribute to the safety analysis of a DC system. To achieve this purpose, we study, first, each DC sub-system (electrical, thermal and network) separately, in order to define their characteristics. Each DC sub-system is a production system and consists of combinations of components that transform entrance supplies (energy for the electrical system, air flow for the thermal one, and packets for the network one) into exits, which can be internet services. Currently the existing safety analysis methods for these kinds of systems are inadequate, because the safety analysis must take into account not only the internal state of each component, but also the different production flows circulating between components. In this thesis, we consider a new modeling methodology called Production Trees (PT) which allows modeling the relationship between the components of a system with a particular attention to the flows circulating between these components.The PT modeling technique allows dealing with one kind of flow at once. Thus its application on the electrical sub-system is suitable, because there is only one kind of flows (the electric current). However, when there are dependencies between sub-systems, as in thermal and network sub-systems, different kinds of flows need to be taken into account, making the application of the PT modeling technique inadequate. Therefore, we extend this technique to deal with dependencies between the different kinds of flows in the DC. Accordingly it is easy to assess the different safety indicators of the global DC system, taking into account the interactions between its sub-systems. Moreover we make some performance statistics. We validate the results of our approach by comparing them to those obtained by a simulation tool that we have implemented based on Queuing Network theory.So far, Production Trees models are not tool supported. Therefore we propose a solution method based on the Probability Distribution of Capacity (PDC) of flows circulating in the DC system. We implement also the PT model using the AltaRica 3.0 modeling language, and use its dedicated stochastic simulator to estimate the reliability indices of the system. This is very important to compare and validate the obtained results with our assessment method. In parallel, we develop a tool which implements the PT solution algorithm with an interactive graphical interface, which allows creating, editing and analyzing PT models. The tool allows also displaying the results, and generates an AltaRica code, which can be subsequently analyzed using the stochastic simulator of AltaRica 3.0 tool.