Internet Cloud Architectures

Today's networks require a deep understanding of applications to optimize networks, efficiently design networks, and meet traffic demands, application heterogeneity, and application requirements. Current application areas include live video streaming and real-time applications, such as those that are named in 5G use cases with automation, disaster recovery, gaming, and Industry 4.0. In this work, we examine an application scenario with live video streaming and parallel real-time requirements in the uplink for disaster recovery. We study the quality of service (QoS) features of a remote-controlled drone. The drone is controlled via a tablet or smartphone while the video from the camera is transmitted from the drone to the user. There are high demands in both the uplink and downlink direction. The contribution of the work is the measurement of the QoS and application parameters for this scenario and the definition of influencing parameters for the application-layer.

Quality of Experience (QoE) of Internet services is affected by human, system, and context influence factors. While most QoE studies so far are focused on system factors only, this work will assess the impact of context factors of video streaming on QoE. As video streaming is mostly consumed from web pages, such as video portals, the investigated test conditions are applied to the web page, which embeds the video player. Therefore, the study of context factors is implicitly conducted within a crowdsourced QoE study. The test conditions considered different page load times, poster image qualities, and displayed advertisements on the web page, which are typical context factors when consuming a video streaming service. The results of the study show that the modification of the context factors on the streaming web page leaves the users' QoE rating unperturbed, which suggests that the investigated context factors have a negligible impact on video streaming QoE, or that the rating task of the subjective QoE study superimposed the context factors.

The fast growth of HTTP video streaming is responsible for a huge amount of traffic over the past few years. Due to the variety and popularity of video content, more and more people are watching videos on the smart TV or on mobile devices. As a result, a potential market is emerging for video providers, which can significantly increase their revenues. In order to offer users a good experience, adaptive video streaming has been introduced to adapt the video quality to the network conditions. Nevertheless, it is still difficult for the network operators to assess the actual video quality on the device of the users and therefore they can not react to improve the service on the network. In this work, we propose a Virtual Network Function (VNF) to monitor the Quality of Experience (QoE) for online video service in the network. To conduct the study, on the one hand, we design a VNF monitoring to measure the video quality and estimate the QoE at the client machine. Our function is placed in two locations nearby and far away from the user to analyze the impact of geographical placement of the VNF on its performance. On the other hand, we set up a local testbed to examine the functional operation and measure the actual video buffer from a client web browser directly to validate the accuracy of the function. Our findings show that with respect to function placement, the VNF has high accuracy in estimating the QoE if it is deployed at the edge network close to the user. However, the VNF does not perform well when it operates far away from the users, e.g., at data centers. These insights help network vendors to more closely monitor the quality of the videos streamed to their customers.

YoMoApp (YouTube Monitoring App) is an Android app to monitor mobile YouTube video streaming on both application- and network-layer. Additionally, it allows to collect subjective Quality of Experience (QoE) feedback of end users. During the development of the app, the stable versions of YoMoApp were already available in the Google Play Store, and the app was downloaded, installed, and used on many devices to monitor streaming sessions. As the app was not advertised in special campaigns or used for dedicated QoE studies, the monitored streaming sessions of this period compose the data set of a large unsupervised field study. The collected data set is evaluated to characterize current mobile YouTube streaming on both application and network layers. Furthermore, the problems and methodology to obtain QoE results from such unsupervised field study are discussed together with the actual QoE results. Correlations between QoE factors are investigated, and the QoE of clusters of similar streaming sessions is analyzed.

The continuous growth of Internet content, applications and services has led to ever more demanding requirements in the networks. To meet this, there have been significant developments in the migration of applications and services towards the cloud. The cloud not only provides applications, but also offers dynamic adaptability, scalability, and optimization with respect to the location of the application by migrating storage and application containers. In particular, cloud services can typically be placed and orchestrated at a variety of locations in the cloud. This demonstration shows a dynamic placement of a cloud service with video streaming and real-time control commands for a remote-controlled drone. We control a drone with the help of a cloud service which, in addition to the control commands, also returns a camera image to the operator. The requirements of this cloud application are groundbreaking. The camera requires high data rates from the drone to the cloud and from the cloud to the operator, so that the image can be transmitted in acceptable quality. For the control of the drone, a short delay must also be given in the direction of the drone so that the control can be carried out in real time.

Edge cloud computing is a trending paradigm, which extends cloud computing by additionally utilizing computing resources at the network edge, e.g., at mobile base stations. Especially personalized services can be instantiated or migrated close to end users, which improves the latency and supports user mobility. However, the placement of the service chains is crucial for the performance of the services and the energy consumption of the edge cloud platform, and appropriate algorithms have to be designed. To support the simulative performance evaluation of such algorithms, EdgeNetworkCloudSim was developed. It is an extension of NetworkCloudSim, and allows to simulate and evaluate the orchestration and consolidation of service chains in an edge network cloud.

In HTTP adaptive streaming (HAS), such as MPEG DASH, the video is split into chunks and is available in different quality levels. If the video chunks are stored or cached on different servers to deal with the high load in the network and the Quality of Experience (QoE) requirements of the users, the problem of content selection arises. In this paper, we evaluate client-side algorithms for dynamically selecting an appropriate content server during DASH video streaming. We present three algorithms with which the DASH client itself can determine the most appropriate server based on client-specific metrics, like actual latency or bandwidth to the content servers. We evaluate and discuss the proposed algorithms with respect to the resulting DASH streaming behavior in terms of buffer levels and quality level selection.

The paradigm of Software as a Service has gained great achievements in the last decade. By transferring computation and storage to the cloud and migrating services to the edge network, users benefit from using demanding services on lightweight devices. However, the user perceived quality of experience (QoE) for these services is facing the challenges of network impairments and the accessibility of users. Unlike a typical PC-based software, the cloud provides users a location-aware, flexible placement of resource for a cost effective service. The geographical placement of content is therefore one of the key factors that affects the user's satisfaction. The closer the content to the user geographically is, the faster it will be delivered to the user that will also increase the user perceived QoE. In this work, we estimate more precisely the QoE for photo loading time in a particular usage of a photo album cloud service with regard to the influence of various parameters. Firstly, we validate a TCP throughput model and use it to calculate the photo loading time from a given photo size and network QoS. Thereafter, we formulate a mapping function to calculate the MOS value from a QoE model adding the output of the TCP model. From this mapping function, we can estimate QoE for photo loading time from a given photo size, its placement and network QoS. Our main contribution is to determine the trade-off between the size of photo and its placement to acquire a high QoE for photo loading time, which is important for the development of a photo album cloud service.

YouTube is one of the most popular and volume-dominant services in today’s Internet, and has changed the Web for ever. Consequently, network operators are forced to consider it in the design, deployment, and optimization of their networks. Taming YouTube requires a good understanding of the complete YouTube stack, from the network streaming service to the application itself. Understanding the interplays between individual YouTube functionalities and their implications for traffic and user Quality of Experience (QoE) becomes paramount nowadays. In this paper we characterize and model the YouTube stack at different layers, going from the generated network traffic to the QoE perceived by the users watching YouTube videos. Firstly, we present a network traffic model for the YouTube flow control mechanism, which permits to understand how YouTube provisions video traffic flows to users. Secondly, we investigate how traffic is consumed at the client side, deriving a simple model for the YouTube application. Thirdly, we analyze the implications for the end user, and present a model for the quality as perceived by them. This model is finally integrated into a system for real time QoE-based YouTube monitoring, highly useful to operators to assess the performance of their networks for provisioning YouTube videos. The central parameter for all the presented models is the buffer level at the YouTube application layer. This paper provides an extensive compendium of objective tools and models for network operators to better understand the YouTube traffic in their networks, to predict the playback behavior of the video player, and to assess how well they do it in the practice with the satisfaction of their customers watching YouTube.

We introduce a concept for client-initiated selection of service location and service quality for improving the Quality of Experience (QoE) of general cloud services. It is loosely based on the HTTP adaptive streaming approach (e.g., MPEG DASH). A manifest file compiled by the cloud service provider specifies the available service locations and qualities, from which the user selects the optimal service instance based on contextual information obtained from client measurements and user preferences. The proposed concept is defined and is implemented in two client-based decision algorithms for improving the QoE of a simple picture gallery cloud service. These decision algorithms are evaluated and their impact on the service delivery is discussed. The evaluation shows that it is possible to improve the service location and quality selection by light-weight client-based algorithms.

YouTube video streaming is one of the most popular and most demanding services in cellular networks. Thus, operators are concerned about the quality of the streaming delivered by their networks and would like to monitor the Quality of Experience (QoE) of the end users. In this work, we conduct a field study of mobile YouTube video streaming, in which both network flow parameters and application-layer streaming parameters were monitored, and present the characteristics of current mobile YouTube streaming. The impact of both approaches is investigated showing that monitoring network parameters is not sufficient to directly infer the resulting QoE. In contrast, the streaming parameters, which can be obtained from application-layer monitoring, show high correlations to the subjectively experienced quality, and thus, are better suited for QoE monitoring.

Free-to-play models, streaming of games and eSports are reasons for online gaming to grow in popularity recently. On the forefront are multiplayer online battle arenas, which gain high popularity by introducing a competitive format that is easy to access and requires cooperation and team play. These games highly rely on fast reaction of the players, which makes latency the key performance indicator of such applications. To obtain low latency, this paper proposes moving game servers close to players towards the edge of the network. The performance of such mechanism highly depends on the geographic distribution of players. By analyzing match histories and statistics, we develop models for the arrival process and location of game requests. This allows us to evaluate the performance of edge server resource migration policies in an event based simulation. Our results show that a high number of edge servers is preferable compared to few larger edge servers to reduce the latency of players. This supports approaches that allow deploying virtual server instances in the back-haul.

The largest share of today`s consumer Internet traffic is video streaming and its demand on content delivery networks is continuously growing. To cope with the increasing demand of video streaming, recent work proposes mitigating end-user equipment to support content delivery at the edge of the network. The throughput of end-user equipment supporting content delivery is limited by the uplink of the users Internet connection.Especially for video streaming insufficient throughput causes the video to stall and affects the Quality of Experience (QoE) of end-users.To prevent video streams from stalling, we consider a tiered caching architecture, which requests higher tier caches to support content delivery, if the uplink throughput drops below a certain threshold. We conduct a simulative performance evaluation of the mechanism to investigate its impact on the QoE of end-users. Our results show that especially if the upload bandwidth of end-user equipment is low the setting of the threshold has a high impact. This can be used by operators to achieve the desired trade-off between QoE and operational cost for cache resources.

To implement improved Quality of Service (QoS) and Quality of Experience (QoE) management for content-heavy services like video streaming, content has to be moved closer to the edge. The concept of information-centric networking (ICN) would be a prospective enabler but is currently not practically feasible yet. We propose a hierarchical caching architecture utilizing caches on home routers to augment existing content delivery network (CDN) infrastructure. This approach can be implemented via Software-defined Networking (SDN) and brings current CDNs closer towards ICN. Based on a simulation study, we confirm that our approach is able to serve content more locally, which results in QoS and QoE benefits for end users as well as inter-domain traffic savings for network operators.

The main goal of this thesis is to define, analyze, and evaluate different orchestration and access principles of highly distributed Internet applications. For this purpose, different cloud application types have to be defined. Thereafter three different ap- plications are modeled, measured, and evaluated during the course of this thesis, namely a cloud based picture gallery, a distributed video streaming application and an interactive image processing application. These applications are supposed to run within an existing OpenStack environment that was developed during an earlier Master’s thesis. In the modeling process of the applications, it has to be taken into account that these applications should be scalable in the horizontal, as well as in the vertical dimension. Scaling in the horizontal dimension corresponds to the duplication or migration of services to different locations on the Internet, while the vertical scaling of an application refers to the scaling of the application scope, the complexity of the application functionality, or the provided content quality for specific users. The remainder of this work is structured as follows: Background information about cloud applications and their architecture is presented in Chapter 2. Thereafter, we discuss work that is related to this thesis in Chapter 3. This is followed by a characterization of different cloud application types in Chapter 4. After that, the orchestration and access principles that are important to this work are briefly de- scribed in Chapter 5. Subsequently, we introduce the cloud applications and their corresponding testbeds that have been developed during this thesis in Chapter 6. These applications are then evaluated under consideration of the previously de- scribed orchestration and access principles in Chapter 7. Finally, we conclude this thesis and try to give an outlook for further work in this area in Chapter 8.

The vast majority of Internet traffic is carried by content delivery networks. A high potential to bring content even closer to consumers and to reduce energy cost is achieved by the nano data center (NaDa) concept that has been proposed in recent work. In this approach the capacity available on small devices such as home gateways is used to support content delivery in an hierarchical caching system. In this work, we present a method to determine the efficiency of hierarchical content delivery networks with limited bandwidth, such as in the NaDa approach, analytically. We analyze the Internet Census Dataset to assess the number of available devices on autonomous system level. We evaluate the potential of hierarchical cache networks with bandwidth constraints to assist content delivery and determine its limits.

YouTube is one of the most popular and volume-dominant services in today’s Internet, and has changed the Web for ever. Consequently, network operators are forced to consider it in the design, deployment, and optimization of their networks. Taming YouTube requires a good understanding of the complete YouTube stack, from the network streaming service to the application itself. Understanding the interplays between individual YouTube functionalities and their implications for traffic and user Quality of Experience (QoE) becomes paramount nowadays. In this paper we characterize and model the YouTube stack at different layers, going from the generated network traffic to the QoE perceived by the users watching YouTube videos. Firstly, we present a network traffic model for the YouTube flow control mechanism, which permits to understand how YouTube provisions video traffic flows to users. Secondly, we investigate how traffic is consumed at the client side, deriving a simple model for the YouTube application. Thirdly, we analyze the implications for the end user, and present a model for the quality as perceived by them. This model is finally integrated into a system for real time QoE-based YouTube monitoring, highly useful to operators to assess the performance of their networks for provisioning YouTube videos. The central parameter for all the presented models is the buffer level at the YouTube application layer. This paper provides an extensive compendium of objective tools and models for network operators to better understand the YouTube traffic in their networks, to predict the playback behavior of the video player, and to assess how well they do it in the practice with the satisfaction of their customers watching YouTube.

We introduce a concept for client-initiated selection of service location and service quality for improving the Quality of Experience (QoE) of general cloud services. It is loosely based on the HTTP adaptive streaming approach (e.g., MPEG DASH). A manifest file compiled by the cloud service provider specifies the available service locations and qualities, from which the user selects the optimal service instance based on contextual information obtained from client measurements and user preferences. The proposed concept is defined and is implemented in two client-based decision algorithms for improving the QoE of a simple picture gallery cloud service. These decision algorithms are evaluated and their impact on the service delivery is discussed. The evaluation shows that it is possible to improve the service location and quality selection by light-weight client-based algorithms.

Online video games and the subjective quality of player inter- actions with them rely on good network conditions. Almost equally important is a good and timely knowledge of such conditions in order to take proper countermeasures to worsening conditions. To this end, many online video games include components for estimating the current network status. In this paper, we examine the accuracy of those estimations for the popular competitive multiplayer game DOTA 2. Our results show that the game client is capable of performing a good estimation of the delay and packet loss, but only after a rather large initial delay.

The main goal of this thesis is to define, analyze, and evaluate different orchestration and access principles of highly distributed Internet applications. For this purpose, different cloud application types have to be defined. Thereafter three different ap- plications are modeled, measured, and evaluated during the course of this thesis, namely a cloud based picture gallery, a distributed video streaming application and an interactive image processing application. These applications are supposed to run within an existing OpenStack environment that was developed during an earlier Master’s thesis. In the modeling process of the applications, it has to be taken into account that these applications should be scalable in the horizontal, as well as in the vertical dimension. Scaling in the horizontal dimension corresponds to the duplication or migration of services to different locations on the Internet, while the vertical scaling of an application refers to the scaling of the application scope, the complexity of the application functionality, or the provided content quality for specific users. The remainder of this work is structured as follows: Background information about cloud applications and their architecture is presented in Chapter 2. Thereafter, we discuss work that is related to this thesis in Chapter 3. This is followed by a characterization of different cloud application types in Chapter 4. After that, the orchestration and access principles that are important to this work are briefly de- scribed in Chapter 5. Subsequently, we introduce the cloud applications and their corresponding testbeds that have been developed during this thesis in Chapter 6. These applications are then evaluated under consideration of the previously de- scribed orchestration and access principles in Chapter 7. Finally, we conclude this thesis and try to give an outlook for further work in this area in Chapter 8.

Durch die steigende Anzahl an Endgeräten im Netzwerk und den damit verbundenen Anstieg an Datenverkehr ist es für den Provider wichtig, die erzeugte Menge an Datenverkehr genau zu kennen. Dies stellt sich durch viele neue Geräte mit neuen Verkehrsmustern jedoch als schwierig heraus, weshalb detaillierte Verkehrsmodelle für jedes Gerät erstellt werden müssen. Auf der anderen Seite profitiert auch der Nutzer von der genauen Verkehrsanalyse. Störungen in von ihm genutzen Services werden schneller erkannt und behoben. In der hier vorgestellten Arbeit soll diese Verkehrsanalyse anhand einer vom Nutzer ferngesteuerten Drohne durchgeführt werden: Ziel der Arbeit ist es mit Hilfe einer am Lehrstuhl vorhandenen Drohne Datenverkehrsmessungen durchzuführen und zu evaluieren. Es soll herausgefunden werden, wie sich die Distanz zum Nutzer oder Hindernisse zwischen dem Nutzer und der Drohne auf den Datendurchsatz beispielsweise bei einer WiFi oder Bluetooth Verbindung auswirkt. Abschließend sollen die gewonnenen Ergebnisse mit Hilfe von vorhandenen Verkehrsmustern verglichen werden und Gemeinsamkeiten sowie Unterschiede herausgefunden und analysiert werden.