Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-04-06T13:08:48Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) A CONTROL-THEORETIC APPROACH FOR ADAPTIVE STREAMING OVER HTT.pdf: 5118913 bytes, checksum: fc08706cd245c8c0de45e4db894cb0a5 (MD5) / Made available in DSpace on 2016-04-06T13:08:48Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) A CONTROL-THEORETIC APPROACH FOR ADAPTIVE STREAMING OVER HTT.pdf: 5118913 bytes, checksum: fc08706cd245c8c0de45e4db894cb0a5 (MD5) Previous issue date: 2015-08-24 / FACEPE / Video applications account for a large percentage of Internet traffic, and their utilization tends to grow. In this scenario, multimedia providers are required to deliver video and/or audio content in a seamless manner, regardless of network conditions and client device capabilities. However, current commercial services are not efficient enough, and the academic proposals do not take full advantage of their potential. To this end, we propose a control-theoretic adaptive streaming over HTTP system, which adapts the video quality to the network conditions in a video session. With this system, we aim at delivering video with the highest Quality of Experience possible, while leveraging the available network resources. Our system is located at the client side, and comprises a control system and a state machine. Together, they select the video level that is requested to the server, a standard HTTP server, which sends video chunks as the client requests them. To select the control system and state machine that best fit our requirements, we perform several experiments with different parameter values combinations. Our parameter selection is accomplished in two steps, the last one resulting in the combination that best fits our needs. Then, we compare our system performance to an existent system in the academia. Our experimental results show the benefits of providing more aggressive decisions on the video quality selection. However, they also show that such aggressiveness may, sometimes, compromise the system performance. Overall, though, our system outperforms the baseline of our experiments. / A computação em nuvem é um paradigma computacional que vem sendo amplamente utilizado ao longo dos últimos anos devido as suas características de provisionamento de recursos de forma escalável, onde o usuário paga apenas por aquilo que consumir. Esse modelo computacional possibilita que diversos serviços sejam ofertados a partir da sua Infraestrutura como Serviço (IaaS - Infrastructure as a Service). Porém, a falha de componentes dos recursos da nuvem é algo bastante comum e que afeta diretamente a disponibilidade dos serviços que os utilizam. Garantir alta disponibilidade em serviços na nuvem é um desafio para provedores, que podem utilizar de técnicas como análise de sensibilidade para identificar gargalos de disponibilidade ou ainda fazer uso de mecanismos de tolerância à falhas para atingir melhores resultados de confiabilidade. Desse modo, esse trabalho tem como proposta realizar a avaliação de disponibilidade em arquiteturas para o provimento do serviço de VoD streaming na nuvem, baseado na plataforma Eucalyptus. O trabalho está dividido em dois estudos de caso. Primeiro, apresentamos uma arquitetura básica, sem mecanismos de redundância. Através de modelagem hierárquica formada por diagramas de blocos de confiabilidade (RBD - Reliability Block Diagram) e cadeias de Markov de tempo contínuo (CTMC - Continuous Time Markov Chain), realizamos a avaliação da disponibilidade para essa arquitetura, e em seguida alcançamos a validação do modelo que representa o funcionamento do serviço. Por fim, aplicamos a técnica de análise de sensibilidade paramétrica para identificar gargalos de disponibilidade na arquitetura básica. As análises seguintes que foram esse estudo de caso observam a disponibilidade do serviço em arquiteturas com redundância no modo warm-stand guiadas por índices de sensibilidade, apresentando ao final uma comparação dos valores de disponibilidade para cada arquitetura obtida a partir de componentes redundantes. O segundo estudo de caso analisa o comportamento da disponibilidade total do serviço a partir da variação no tempo de ativação dos mecanismos redundantes. Nossos resultados expressaram que a adoção de componentes redundantes no modo warm-stand na infraestrutura da nuvem proporciona ganhos de disponibilidade no serviço analisado. Também observamos que a variação do tempo de ativação do componente de maior confiabilidade tem maior impacto na disponibilidade total do serviço de VoD streaming.

Användandet av strömmande medier har ökat explosionsartat under de senaste åren. Oavsett om det sker märkbart för användaren eller inte använder sig gemene man i princip dagligen av någon form av tjänst som distribueras via strömmande medier över Internet. Det kan röra sig om allt från att lyssna på webbradio till att titta på videoklipp från nyhetsportaler eller sportsidor på Internet. I denna rapports första del görs en kartläggning över strömmande medier rent generellt. Den läsekrets som inte är särskilt insatt i ämnet får reda på vad strömmande medier är, vilka aktörer som finns på marknaden, möjliga distributionssätt, historik med mera. Fortsättningsvis kommer rapporten att behandla hur införandet av strömmande medier går till i teorin: hur distributionskedjan ser ut, vilka krav som konsument och producent ställer och hur man försöker att lösa problem som uppstår. Det senare leder vidare till en genomgång av bland annat streamingprotokoll, bredbandstekniker och komprimeringsstandarder för ljud och bild. I rapportens senare del ges exempel på hur strömmande medier införs i praktiken. Här läggs speciellt fokus på användningsområdet sport och direktsändningar av sportevenemang. Konkreta exempel beskrivs och vi får även reda på hur det går till att genomföra en direktsänd sportsändning via strömmande medier. Som avslutning diskuteras tänkbara framtidscenarion och en utredning av en unikInternetsändning av en elitishockeymatch görs.

Målen med denna studie är att implementera en testmiljö för streaming telemetry samt jämföra två alternativ för att möjliggöra realtidsanalys av det insamlade datat. Dessa två alternativ är Python-biblioteken PyKafka och Confluent-Kafka-Python. Bedömningskritierna för jämförselsen var dokumentation, kodmängd och minnesanvändning. Testmiljön för streaming telemetry använder en router med Cisco IOS XR programvara som skickar data till en Cisco Pipeline collector, som vidare sänder datat till ett Kafka-kluster. Jämförelsen av Python-biblioteken utfördes med språket Python. Resultaten av jämförelsen visade att båda biblioteken hade välskriven dokumentation och liten skillnad i kodmängd, dock använde Confluent-Kafka-Python mindre minne. Studien visar att streaming telemetry med realtidsanalys kan fungera bra som ett komplement till eller en ersättning av SNMP. Studien rekommenderar användning av Confluent-Kafka-Python för implementering i produktionsmiljöer med ett stort antal nätverksenheter med tanke på den lägre minnesanvändningen. / The goals of this study are to implement a test environment for streaming telemetry and compare two alternatives for analysing the collected data in realtime. The two alternatives are the Python libraries PyKafka and Confluent-Kafka-Python. The comparison focused mainly on three areas, these being documentation, amount of code and memory usage. The test environment for streaming telemetry was set up with a router running IOS XR software that is sending data to a Cisco Pipeline collector, which in turn sends data to a Kafka-cluster. The comparison of the two libraries for interfacing with the cluster was made with the language Python. The results of the comparison showed that both libraries had well-written documentation and showed a negligible difference in amount of code. The memory usage was considerably lower with the Confluent-Kafka-Python library. The study shows that streaming telemetry together with real-time analysis makes a good complement to or a replacement of SNMP. The study further recommends the use of Confluent-Kafka-Python in real-world implementations of streaming telemetry, particularly in large networks with a large amount of devices.

Accelerated advancements in technology, the Internet of Things, and cloud computing have spurred an emergence of unstructured data that is contributing to rapid growth in data volumes. No human can manage to keep up with monitoring and analyzing these unbounded data streams and thus predictive and analytic tools are needed. By leveraging machine learning this data can be converted into insights which are enabling datadriven decisions that can drastically accelerate innovation, improve user experience, and drive operational efficiency. The purpose of this thesis is to design and implement a system for real-time outlier detection using unbounded data streams and machine learning. Traditionally, this is accomplished by using alarm-thresholds on important system metrics. Yet, a static threshold cannot account for changes in trends and seasonality, changes in the system, or an increased system load. Thus, the intention is to leverage machine learning to instead look for deviations in the behavior of the data not caused by natural changes but by malfunctions. The use-case driving the thesis forward is real-time outlier detection in a Content Delivery Network (CDN). The input data includes Http-error messages received by clients, and contextual information like region, cache domains, and error codes, to provide tailormade predictions accounting for the trends in the data. The outlier detection system consists of a data collection pipeline leveraging the technique of stream processing, a MiniBatchKMeans clustering model that provides online clustering of incoming data according to their similar characteristics, and an LSTM AutoEncoder that accounts for temporal nature of the data and detects outlier data points in the clusters. An important finding is that an outlier is defined as an abnormal amount of outlier data points all originating from the same cluster, not a single outlier data point. Thus, the alerting system will be implementing an outlier percentage threshold. The experimental results show that an outlier is detected within one minute from a cache break-down. This triggers an alert to the system owners, containing graphs of the clustered data to narrow down the search area of the cause to enable preventive action towards the prominent incident. Further results show that within 2 minutes from fixing the cause the system will provide feedback that the actions taken were successful. Considering the real-time requirements of the CDN environment, it is concluded that the short delay for detection is indeed real-time. Proving that machine learning is indeed able to detect outliers in unbounded data streams in a real-time manner. Further analysis shows that the system is more accurate during peakhours when more data is in circulation than during none peak-hours, despite the temporal LSTM layers. Presumably, an effect from the model needing to train on more data to better account for seasonality and trends. Future work necessary to put the outlier detection system in production thus includes more training to improve accuracy and correctness. Furthermore, one could consider implementing necessary functionality for a production environment and possibly adding enhancing features that can automatically avert incidents detected and handle the causes of them.

La presente investigación tiene la finalidad de explicar cómo el E-Service Quality y la intención de uso que influyen, como variables independientes, en la E-Satisfaction denominada variable dependiente en las plataformas de servicio de streaming. Con ello, se busca comprender de mejor manera la satisfacción de los clientes durante el uso del servicio. Así mismo, el estudio realizará una investigación exhaustiva sobre las características principales de las variables involucradas, con el fin de poder comprender de manera eficaz dicho procedimiento. Por ello, el muestreo será no probabilístico debido a la población infinita del presente estudio y a los parámetros que se pueden establecer. La muestra estará conformada por hombres y mujeres con edades entre los 18 y 40 años y que utilicen de manera frecuente los servicios de las plataformas de streaming de TV. Por último, en el testeo estadístico se aplicarán correlaciones y regresiones lineales múltiples con el objetivo de establecer relaciones entre las variables analizadas. / The present research aims to explain how the E-Service Quality and the intention of use that influence, as independent variables, the E-Satisfaction called dependent variable in streaming service platforms. With this, we seek to understand the customer satisfaction during the use of the service better. At the same time, the study will carry out an exhaustive investigation on the main characteristics of the variables involved, to be able to understand this procedure effectively. Therefore, the sampling will be non-probabilistic due to the infinite population of the present study and the parameters that can be established. The sample will come from men and women between the ages of 18 and 40 and who use the services of the streaming TV platforms frequently. Finally, in the statistical testing, multiple linear correlations and regressions will be applied in order to establish relationships between the analyzed variables. / Trabajo de investigación

Dans cette thèse nous étudions l'impatience des usagers dans les réseaux mobile et nous quantifions son impact sur la performance du système en présence d'usagers téléchargeant des données dans lequel nous développons deux expressions approximatives de la distribution stationnaire du systéme : un modèle agrégé et un modèle détaillé et nous montrons qu'ils sont très proches du modèle exact. Nous étudions la mobilité de l'usager téléchargeant des données et pouvant s'impatienter, et nous quantifions son impact sur la performance des réseaux mobile. Nous considérons le cas de la mobilité due à l'impatience et le cas de la mobilité spontanée des usagers tout en considérant la mobilité intra et inter cellulaire. Nous étudions également l'impatience de l'usager qui regarde une vidéo streaming durant les phases de pré-chargement de la vidéo et de mise en tampon pendant que la vidéo est arrêtée dès le début de la mise en tampon. Nous étudions à la fin un système constitué d'usagers pouvant s'impatienter, qui est sous contrôle d'un gestionnaire de système, qui à chaque instant de décisions, choisit une action à exécuter dans le but d'optimiser la performance définie du système. Nous considérons un système dans lequel les usagers arrivent dans le système à des différents instants et le quittent après la fin de leurs transferts de données, ou plus tôt à l'expiration de leurs durées de patience. Les applications numériques et les simulations nous ont permis de fournir divers métriques de performance telles que le nombre moyen d'usagers, la proportion d'usagers impatients qui quittent le système avant la fin de leurs transferts de fichiers, le débit, la probabilité d'impatience en tenant compte de la localisation de l'usager dans la cellule, la probabilité d'être impatient durant les phases de mise en tampon, la probabilité de mise en tampon lors d'une séance de vidéo streaming, la politique optimale, la taille limite optimale du système dans le but d'optimiser la performance définie du système, etc... / In this thesis we study user impatience and quantify its impact on the performance of mobile networks in the presence of elastic user in which we develop two approximate expressions for the stationary distribution of the system: an aggregate one and a detailed one and show their closeness to the exact model. We study mobility of elastic user who may be impatient and quantify its impact on system performance in mobile networks. We consider the case of mobility due to impatience during the prefetching and the re-buffering phases when starvation happens. We finally study a system with impatient users controlled by a system manager who has to choose at each decision epoch an action to make in order to optimize the defined system performance. We consider a system in which users come to the system at different time instants and leave it after a finite duration, either after completion of their data transfers or earlier, at the expiry of some patience duration. Numerical analysis and simulations allow us to derive several performance metrics such as mean number of users, the proportion of impatient users who quit the system before completing their file transfers, the throughput, the probability of starvation, the optimal system size in order to optimize the defined performance of the system, etc...

Background：Live streaming shopping is becoming a mainstream method in China since the livestreaming feature has been applied to e-commerce. Live streaming commerce reflects theinterdependence between consumers and live streamers. Past studies have found that the ability,social presence and reputation of live streamers influence consumers' purchase intentions.However, there is a gap in how consumers' purchase intentions are affected by negative incidentsand a lack of qualitative perspective. Purpose：To explore consumers' understanding of their purchase intentions in relation to liveshopping streamers, in the context of negative incidents with streamers. Methodology：This study used qualitative research as the research method and abductivereasoning to conduct the data analysis. This study used semi-structured interviews and developedinterview questions based on the content of the theory chapter. The researchers usednon-probability sampling and general purpose sampling as the sampling strategy. The requireddata were obtained from 12 Chinese interviewees who had experience with live shopping. Theobtained data were then coded using grounded theory as the analysis method for qualitative data,which helped the researchers understand the thoughts of the interviewees. The final attempt is todevelop a theory that can explain the phenomenon. Moreover, the methods chapter explores thequality of the research in this study, as well as possible ethical and social issues. Findings：This paper identified three consumer needs (utilitarian, pleasure-seeking andemotional needs) and factors associated with streamers (attitude, streaming style, impact,characteristics and ability) that influence purchase intentions in the presence of negative streamerincidents. The paper also found how consumers' purchase intentions were influenced by thesefactors in different negative incident situations. Conclusion：This study obtained consumers' understanding of the relationship between theirpurchase intentions and streamers. Consumers exhibit three different levels of purchase3intention: willing to buy, wait-and-see (unsure of purchase intention) and no purchase intention.In the case of negative incidents, consumers' different purchase intentions are influenceddifferently by their own needs and streamer factors.

Följande studie undersöker musikkonsumenternas behov i relation till streamingtjänsterna, med utgångspunkt i det krig som pågår mellan streamingtjänsterna, samt i de tre aktuellastreamingtjänsterna Spotify, Deezer och Apple Music. Studien undersöker musikkonsumentens (tolv till 35 år gamla) omedvetna behov som musikstreamingtjänsten ska uppfylla, samt hur förhållandet ser ut mellan de här behoven och musikkonsumentens val av streamingtjänst. Studien har genom en kvalitativ metod, där fokusgrupper och en semistrukturerad intervju legat till grund för empiriinsamlingen, kommit fram till slutsatsen att musikkonsumenten har två övergripande omedvetna behov: personifiering och igenkänning i tjänsten, samt behovet av att streamingtjänsten ska göra det möjligt för musikkonsumenten att leva i streamingen, i flow med den streamade musiken. Förhållandet mellan de omedvetna behoven och valet av streamingtjänst innebär övergripande att musikkonsumenten använder den streamingtjänst som tillfredsställer de omedvetna behoven, men inte byter tjänst med utgångspunkt i sina omedvetna behov, eftersom konsumenten inte själv känner till behoven. / The following study examines the needs music consumers have in relation to the streamingservices, based on the war going on on the streaming market, as well as in the three currentstreaming services Spotify, Deezer and Apple Music. The study examines the music consumers (twelve to 35 years old) unconscious needs that the music streaming service must meet, and how the relationship is shaped between these needs and the music consumer's choice of streaming service. The study has through a qualitative approach, where focus groups and one semi-structured interview formed the basis of empirical data collection, come to the conclusion that the music consumer has two unconscious needs: personalisation and self-recognition in the streaming service, as well as the need for the streaming service to allow the music consumers to live in the stream, in flow with the stream of music. The relationship between these unconscious needs and the choice of streaming service, shows that the music consumer is using the streaming service that satisfies their unconscious needs, but does not replace the service on the basis of their unconscious needs since the consumer do not know these needs themselves.

[pt] A presente pesquisa busca analisar transformações na televisão, a partir da inserção de telenovelas inéditas em plataformas de streaming. Para isso, exploramos mudanças, permanências e adaptações que os serviços de vídeo por demanda estão realizando para a projeção do principal formato narrativo brasileiro em outros modelos de exibição e consumo. O estudo também discute as narrativas contemporâneas serializadas, as características clássicas da telenovela no Brasil e sua influência cultural, as transformações na espectatorialidade e no modelo de negócios do audiovisual dentro do contexto hipermidiático. A investigação está ancorada metodologicamente na teoria das mediações proposta por Martín-Barbero e em uma pesquisa de recepção com telespectadores de Todas as flores – segunda novela brasileira do streaming e primeira totalmente produzida para o Globoplay. Esta análise qualitativa de novas práticas de consumo da teledramaturgia nacional busca uma maior compreensão dos caminhos dos conteúdos ficcionais e das experiências participativas diante das transformações tecnológicas que transcendem as fronteiras tanto do formato quanto do dispositivo original. / [en] This research aims to analyze the transformations in television, through the insertion of new telenovelas on streaming platforms. To do so, we explored changes, continuations and adaptations that video on demand services is doing to project the main Brazilian narrative format into other exhibition and consumption models. The study also discusses contemporary serialized narratives, the classic characteristics of telenovelas in Brazil and their cultural influence, the transformations in spectatorship and the audiovisual business model within the hypermedia context. The investigation is methodologically anchored by the map of mediations proposed by Martin-Barbero and in a reception survey with viewers of Todas as Flores – the second Brazilian telenovela on streaming and the first original produced for Globoplay. This analysis of new consumption practices of national television drama seeks a greater understanding of the paths of fictional content and participatory experiences in the face of technological transformations that transcend the boundaries of both the format and the original device.

[pt] Este trabalho teve como objetivo analisar a abordagem da temática da violência contra a mulher em produções audiovisuais, especialmente em plataformas de streaming. A pesquisa explora como essas obras ajudam a ampliar os debates sociais e encorajar o público a se engajar, usando a narrativa como uma ferramenta poderosa. O objeto analisado foi a série de TV Bom dia, Verônica, da Netflix, que recebeu prêmios e aumentou as vendas do livro relacionado à trama. Durante uma entrevista com o autor, Raphael Montes, foi destacado que a série incluiu mensagens de apoio à denúncia de violência, resultando em muitas ligações para uma central de denúncias parceira da obra. A pesquisa também ressaltou que a série refletiu experiências reais, como o medo de denunciar, enfrentado por muitas mulheres. Em resumo, o estudo mostra como essa produção não apenas influenciou o meio cultural, mas também teve impacto real, aumentando a conscientização e apoiando vítimas de violência contra a mulher. / [en] This work the approach to the theme of violence against women in audiovisual audiovisual productions, especially on streaming platforms. The research explores how these works help to broaden social debates and encourage audiences to engage, using narrative as a powerful tool. The object analyzed was the Netflix TV series Bom dia, Verônica, which won awards and increased sales of the book related to the plot. During an interview with the author, Raphael Montes, it was pointed out that the series included messages of support for reporting violence, resulting in many calls to a reporting center that is a partner of the work. The research also highlighted that the series reflected real experiences, such as the fear of reporting faced by many women. In short, the study shows how this production not only influenced the cultural milieu, but also had a real impact, raising awareness and supporting victims of violence against women.