IT equipment has been estimated to be responsible for 2% of global CO2 emissions and data centers are responsible for 1.2% of U.S. energy consumption. With the large quantity of high quality digital content available on the Internet the energy demands and environmental impact of the data centers must be addressed. The use of peer-to-peer technologies, such as BitTorrent, to distribute legal content to consumers is actively being explored as a means of reducing both file download times and the energy consumption of data centers. This approach pushes the energy use out of the data centers and into the homes of content consumers (who are also then content distributors). The current BitTorrent protocol requires that clients must be fully powered-on to be participating members in a swarm. In this thesis, an extension to the BitTorrent protocol that utilizes long-lived knowledge of sleeping peers to enable clients to sleep when not actively distributing content yet remain responsive swarm members is developed. New peer states and events required for the protocol extension, the implementation the new protocol in a simulation environment, and the implementation of the protocol extension in a real client are described. Experiments on a simulated swarm of 51 peers transferring a 1 GB and a real swarm of 11 peers transfer- ring a 100 MB file were run. To validate the simulation a simulated swarm of 11 peers transferring a 100 MB file is compared to the real swarm of 11 peers. The results of standard BitTorrent are compared to the new Green BitTorrent by examining download times, sleep time, and awake time. The results of the experiment show significant energy savings are possible with only a small penalty in download time. Energy savings of up to 75% are shown with download time increases as little as 10%. These energy savings could equate to over $1 billion dollars per year in the US alone if Green BitTorrent is used instead of standard BitTorrent for future rollouts of legal distribution systems.

Due to the ongoing climate crisis, reducing waste and carbon emissions has become hot topic in many fields of study. Cloud data centers contribute a large portion to the world’s energy consumption. In this work, methodologies are developed using machine learning algorithms to improve prediction of the energy consumption of a container in a data center. The goal is to share this information with the user ahead of time, so that the same can make educated decisions about their environmental footprint.This work differentiates itself in its sole focus on optimizing prediction, as opposed to other approaches in the field where energy modeling and prediction has been studied as a means to building advanced scheduling policies in data centers. In this thesis, a qualitative comparison between various machine learning approaches to energy modeling and prediction is put forward. These approaches include Linear, Polynomial Linear and Polynomial Random Forest Regression as well as a Genetic Algorithm, LSTM Neural Networks and Reinforcement Learning. The best results were obtained using the Polynomial Random Forest Regression, which produced a Mean Absolute Error of of 26.48% when run against data center metrics gathered after the model was built. This prediction engine was then integrated into a Proof of Concept application as an educative tool to estimate what metrics of a cloud job have what impact on the container power consumption.

Les travaux de cette thèse portent sur l'optimisation des flux énergétiques et informatiques dans un réseau intelligent ayant pour but d'alimenter un centre de calcul via des énergies renouvelables. Dans cette thèse sont traités les problèmes liés à la mise en commun des informations de types énergétique et informatique dans une contrainte de réactivité forte à travers la création d'une architecture pour un réseau intelligent. La modélisation d'un tel réseau doit permettre la prise de décision de manière dynamique et autonome. L'objectif de cette modélisation, via un réseau intelligent, est l'optimisation des ressources renouvelables afin de diminuer l'empreinte écologique. / The work of this thesis deals with the optimization of energy and computer flows in an intelligent network aiming to supply a data center via renewable energies. In this thesis are treated the problems related to the pooling of energy and computer information in a strong reactivity constraint through the creation of an architecture for an intelligent network. The modeling of such a network must allow the decision making in a dynamic and autonomous way. The objective of this modeling, via an intelligent network, is the optimization of renewable resources in order to reduce the ecological footprint.

No description available.

Sub-ambient direct-on-chip liquid cooling is an emerging technology in the data center industry. The risk of an electrically conductive liquid leaking out to the electrical components and damaging the servers has been the major factor in holding back the use of liquid cooling historically. This technology effectively removes that risk. A direct-on-chip liquid cooling system, where average system pressure and average CPU temperatures can be fixed for a range of server computing loads and coolant supply temperatures for data center servers has been designed and constructed. This has been used to determine what impact pressure has on a small-scale liquid cooled server system in terms of CPU power consumption and CPU temperatures. The cooling system was only able to work with one server connected. Experiments with different values for the CPU temperature setpoint, coolant supply temperature setpoint, server computational load, and server pressure were executed to verify that the system works as intended. Applying a range of CPU computing loads works well, maintaining fixed average CPU temperatures works, with differences between the CPUs at higher temperatures and failure to reach average CPU temperatures when the difference between these and the coolant supply temperature is small. Maintaining fixed average pressure before the server works well, while pressure after the server is heavily affected by coolant flow. However, this effect is not seen as important for the experimental goals of the thesis. Maintaining a fixed coolant supply temperature works well with some slow fluctuations around the setpoint. No noticeable effects from pressure on CPU power consumption and CPU temperatures were seen. However, lower flow resistance was seen by the circulating pump when negative system pressure was lower which implies that less pump energy is needed to pump at lower negative pressure. The pressure was not in the region where the coolant could phase change during the experiments.

Due to constantly increasing amount of data, the need for more efficient data center management solutions has increased dramatically. Approximately 40% of the costs for data centers is associated with cooling, making temperature management of data centers vital for data center profitability, performance, and sustainability. Current data center hardware management software lack a visual and contextual approach to data center monitoring, overlooking the hierarchical and spatiotemporal data structures of data center data in its design. This study compared two potential data center temperature visualizations — 3D visualization in virtual reality (VR) and 2D visualization on 2D screen — in terms of time to task completion, accuracy rate, and user satisfaction. Results of a within-subject user study with 13 data center specialists indicated that users perceived three-dimensional data center racks and devices more efficiently in VR than in a 2D visualization, whereas a two-dimensional graph was interpreted more efficiently and accurately on a 2D screen. The user satisfaction of both implemented visualizations scored over 80 in a System Usability Scale (SUS) survey, showing that the implemented visualizations have significant potential to improve data center temperature management. / På grund av den ständigt ökande mängden av data, har behovet till effektivare datacenterhanteringslösningar ökat dramatiskt. Cirka 40% av kostnaderna för datacentrar används till kylning, vilket gör temperaturhanteringen till en kritisk del av datacentrets lönsamhet, prestanda och hållbarhet. Nuvarande datacenterhanteringsprogramvaror saknar visuella och kontextuella tillvägagångssätt för datacenterövervakning och förbiser de hierarkiska och spatiotemporala datastrukturerna för datacenterdata i programvarudesign. Denna studie jämförde två potentiella datacentertemperaturvisualiseringar — en tredimensionell visualisering i virtuell verklighet (VV) och en tvådimensionell visualisering på en 2D skärm — i jämförelsen beaktas tid till uppgiftens slutförande, antalet riktiga svar och tillfredsställelse av användaren. Resultatet av användarstudien med 13 datacenterspecialister antydde att användare uppfattar tredimensionellaelektronikrack och enheter snabbare i VV än med 2D-visualisering, medan en tvådimensionell graf tolkas snabbare och noggrannare på en 2D skärm. Användartillfredsställelse av båda visualiseringarna fick över 80 poäng i SUS mätningen, vilket antyder att de genomförda visualiseringarna har en stor potential för att förbättra datacentertemperaturhanteringen.

Data is increasingly stored in data lakes, vast immutable object stores that can be accessed from anywhere in the data center. By providing low cost and scalable storage, today immutable object-storage based data lakes are used by a wide range of applications with diverse access patterns. Unfortunately, performance can suffer for applications that do not match the access patterns for which the data lake was designed. Moreover, in many of today's (non-hyperscale) data centers, limited bisectional bandwidth will limit data lake performance. Today many computer clusters integrate caches both to address the mismatch between application performance requirements and the capabilities of the shared data lake, and to reduce the demand on the data center network. However, per-cluster caching; i) means the expensive cache resources cannot be shifted between clusters based on demand, ii) makes sharing expensive because data accessed by multiple clusters is independently cached by each of them, and iii) makes it difficult for clusters to grow and shrink if their servers are being used to cache storage. In this dissertation, we present two novel data-center wide cooperative cache architectures, Datacenter-Data-Delivery Network (D3N) and Directory-Based Datacenter-Data-Delivery Network (D4N) that are designed to be part of the data lake itself rather than part of the computer clusters that use it. D3N and D4N distribute caches across the data center to enable data sharing and elasticity of cache resources where requests are transparently directed to nearby cache nodes. They dynamically adapt to changes in access patterns and accelerate workloads while providing the same consistency, trust, availability, and resilience guarantees as the underlying data lake. We nd that exploiting the immutability of object stores significantly reduces the complexity and provides opportunities for cache management strategies that were not feasible for previous cooperative cache systems for le or block-based storage. D3N is a multi-layer cooperative cache that targets workloads with large read-only datasets like big data analytics. It is designed to be easily integrated into existing data lakes with only limited support for write caching of intermediate data, and avoiding any global state by, for example, using consistent hashing for locating blocks and making all caching decisions based purely on local information. Our prototype is performant enough to fully exploit the (5 GB/s read) SSDs and (40, Gbit/s) NICs in our system and improve the runtime of realistic workloads by up to 3x. The simplicity of D3N has enabled us, in collaboration with industry partners, to upstream the two-layer version of D3N into the existing code base of the Ceph object store as a new experimental feature, making it available to the many data lakes around the world based on Ceph. D4N is a directory-based cooperative cache that provides a reliable write tier and a distributed directory that maintains a global state. It explores the use of global state to implement more sophisticated cache management policies and enables application-specific tuning of caching policies to support a wider range of applications than D3N. In contrast to previous cache systems that implement their own mechanism for maintaining dirty data redundantly, D4N re-uses the existing data lake (Ceph) software for implementing a write tier and exploits the semantics of immutable objects to move aged objects to the shared data lake. This design greatly reduces the barrier to adoption and enables D4N to take advantage of sophisticated data lake features such as erasure coding. We demonstrate that D4N is performant enough to saturate the bandwidth of the SSDs, and it automatically adapts replication to the working set of the demands and outperforms the state of art cluster cache Alluxio. While it will be substantially more complicated to integrate the D4N prototype into production quality code that can be adopted by the community, these results are compelling enough that our partners are starting that effort. D3N and D4N demonstrate that cooperative caching techniques, originally designed for file systems, can be employed to integrate caching into today’s immutable object-based data lakes. We find that the properties of immutable object storage greatly simplify the adoption of these techniques, and enable integration of caching in a fashion that enables re-use of existing battle tested software; greatly reducing the barrier of adoption. In integrating the caching in the data lake, and not the compute cluster, this research opens the door to efficient data center wide sharing of data and resources.

Made available in DSpace on 2016-12-12T20:22:53Z (GMT). No. of bitstreams: 1 Mauro Hinz.pdf: 2658972 bytes, checksum: 50ee82c291499d5ddc390671e05329d4 (MD5) Previous issue date: 2015-09-29 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / The outsourcing of computing services has been through constant evolutions in the past years, due to the increase of demand for computing resources. Accordingly, data centers are the main suppliers of computing service and cloud-based computing services provide a new paradigm for the offer and consumption of these computing resources. A substantial motivator for using cloud computing is its pricing model, which enables to charge the customer only for the resources he used, thus adopting a pay-as-you-use cost model. Among cloud-based computing services, the service type Infrastructure-as-a-Service (IaaS) is the one mostly used by companies that would like to outsource their computing infrastructure. The IaaS service, in most cases, is offered through virtual machines. This paper revisits the cost models used by data centers and analyses the costs of supply of virtual machines based on IaaS. This analysis identifies that electricity represents a considerable portion of this cost and that much of the consumption comes from the use of processors in virtual machines, and that this aspect is not considered in the identified cost models. This paper describes the Virtual Power Model, a cost model based on energy consumption of the processor in cloud-based, virtual machines in IaaS. The model is based on the assumptions of energy consumption vs. processing load, among others, which are proven through experiments in a test environment of a small data center. As a result, the Virtual Power Model proves itself as a fairer pricing model for the consumed resources than the identified models. Finally, a case study is performed to compare the costs charged to a client using the cost model of Amazon for the AWS EC2 service and the same service charged using the Virtual Power Model. / A terceirização dos serviços de computação tem passado por evoluções constantes nos últimos anos em função do contínuo aumento na demanda por recursos computacionais. Neste sentido, os data centers são os principais fornecedores de serviço de computação e os serviços de computação em nuvem proporcionam um novo paradigma na oferta e consumo desses recursos computacionais. Um considerável motivador do uso das nuvens computacionais é o seu modelo de tarifação que possibilita a cobrança do cliente somente dos recursos que ele utilizou, adotando um modelo de custo do tipo pay-as-you-use. Dentre os serviços de computação em nuvem, o serviço do tipo IaaS (Infrastructure-as-a-Service) é um dos mais utilizados por empresas que desejam terceirizar a sua infraestrutura computacional. O serviço de IaaS, na grande maioria dos casos, é ofertado através de instâncias de máquinas virtuais. O presente trabalho revisita os modelos de custos empregados em data centers analisando a formação dos custos no fornecimento de máquina virtuais em nuvens baseadas em IaaS. Com base nesta análise identificasse que a energia elétrica possui uma parcela considerável deste custo e que boa parte deste consumo é proveniente do uso de processadores pelas máquinas virtuais, e que esse aspecto não é considerado nos modelos de custos identificados. Este trabalho descreve o Modelo Virtual Power, um modelo de custo baseado no consumo de energia do processador por máquina virtual em nuvens IaaS. A constituição do modelo está baseada nas premissas de consumo de energia vs. carga de processamento, entre outros, que são comprovados através de experimentação em um ambiente de testes em um data center de pequeno porte. Como resultado o Modelo Virtual Power mostra-se mais justo na precificação dos recursos consumidos do que os modelos identificados. Por fim, é realizado um estudo de caso comparando os custos tarifado a um cliente empregando o modelo de custo da Amazon no serviço AWS EC2 e o mesmo serviço tarifado utilizando o Modelo Virtual Power.

With the growing business impact of distributed big data analytics jobs, it has become crucial to optimize their execution and resource consumption. In most cases, such jobs consist of multiple sub-entities called tasks and are executed online in a large shared distributed computing system. The ability to accurately estimate runtime properties and coordinate execution of sub-entities of a job allows a scheduler to efficiently schedule jobs for optimal scheduling. This thesis presents the first study that highlights spatial dimension, an inherent property of distributed jobs, and underscores its importance in efficient cluster job scheduling. We develop two new classes of spatial dimension based algorithms to<br>address the two primary challenges of cluster scheduling. First, we propose, validate, and design two complete systems that employ learning algorithms exploiting spatial dimension. We demonstrate high similarity in runtime properties between sub-entities of the same job by detailed trace analysis on four different industrial cluster traces. We identify design challenges and propose principles for a sampling based learning system for two examples, first for a coflow scheduler, and second for a cluster job scheduler.<br>We also propose, design, and demonstrate the effectiveness of new multi-task scheduling algorithms based on effective synchronization across the spatial dimension. We underline and validate by experimental analysis the importance of synchronization between sub-entities (flows, tasks) of a distributed entity (coflow, data analytics jobs) for its efficient execution. We also highlight that by not considering sibling sub-entities when scheduling something it may also lead to sub-optimal overall cluster performance. We propose, design, and implement a full coflow scheduler based on these assertions.

No description available.