CAPÍTULO IV. LA SOLUCIÓN EXTRAJUDICIAL DE LITIGIOS DE CONSUMO
A. LA MEDIACIÓN
This dissertation presented, in Chapter 2, the technological setting for this thesis: a literature review of online video delivery with an emphasis on video streaming over HTTP and MPEG- DASH. Section 6.1.1 presents a comparison of the proposed DPEA solution with other solutions in this area. Additionally, Chapter 3 presented a literature review in the area of Web-based learning systems with a focus on adaptive Personalised Learning (PL) systems. Such systems, including Adaptive Hypermedia systems were investigated to identify issues relating to learning content adaptation to the delivery context. Insights and a brief comparison of DPEA with the related solutions in the area are presented in Section 6.1.2.
6.1.1 MPEG-DASH Setting
Video streaming approaches have shifted from UDP-based to TCP-based in recent years. Most existing HTTP/TCP-based solutions are proprietary (e.g. Adobe HDS [99], Apple HLS [100], Microsoft Smooth Streaming [101]). MPEG-DASH [12] is an international standard for describing multi-rate encoded multimedia for adaptive HTTP streaming. Client players dynamically choose the quality (bitrate) for segments of a DASH media presentation to request the best match to estimated current network dynamics and/or to available device resources. DASH-based content is growing increasingly prevalent, where the quantity of free and commercially available videos is expanding rapidly. The DPEA architecture proposed in this research enhances DASH video distribution in a campus setting by utilising best performing local and remote hosts.
dPOAA component of DPEA evaluates remote servers based on their historic performance in terms of the measured throughput and RTT of the link to the server. This rating is used for remote server selection when the requested video resides on multiple servers.
Video content is typically delivered by CDNs which host videos at a number of servers. Distributed DASH datasets such as [124] provide identical DASH content on multiple sites. The standard supports provision of alternate base URLs through the BaseURL element at any level
when identical segments are accessible at multiple locations. Section 5.6.5 (Alternative base URLs) of the standard states: “In the absence of other criteria, the DASH Client may use the first BaseURL element as ‘base URI’. The DASH Client may use base URLs provided in the BaseURL element as ‘base URI’ and may implement any suitable algorithm to determine which URLs it uses for requests.” [13, p. 66]. Accordingly, the first challenge after “retrieving an MPD with multiple BaseURLs is determining with which BaseURL to start a DASH session. As the BaseURL does not have any metrics associated (some text omitted) it is up to client implementation to decide the location of the first segments to be downloaded.” [124, p. 134]. The same source stipulates that determining the best BaseURL may influence the initial delay. While the standard supports specification of multiple hosting servers, it does not propose a selection algorithm. To the best of our knowledge, there are no other DASH-based solutions that provide intelligent remote host selection based on statistical estimators.
Server selection strategies are typically deployed by content providers (e.g. within a CDN) to reduce cost and to improve the end-user experience through load balancing. While they utilise proprietary algorithms, studies reveal the algorithms applied by content providers are geographically (locality) aware (e.g. YouTube [269]) and mainly static in nature (e.g. Netflix [107]). Proposed solutions, such as the Control Plane framework [136] allocate CDNs based on global knowledge of delivery network (CDN performance, client activity, etc.).
It can be argued that clients are ideally positioned to observe local network performance and consequently to react promptly to network dynamics, so dPOAA chooses servers based on their historical performance observed from a client’s perspective without any input from the hosting server. A client-based approach to dynamic CDN selection was explored in [149], where multiple dynamic probes were used to identify the best performing CDN at session startup. In contrast, dPOAA selects servers based on historic readings without incurring additional probing traffic.
DAV components of DPEA utilise locally available content through modification of the MPD file provided to the video requesters. Here, DAV is compared to solutions that propose use of content residing on peers and to systems that centrally utilise client provided information. The peer-assisted DASH system (pDASH) [117] was described in Section 2.4.12. pDASH, similar to DAV, modifies MPD files. In the pDASH setting modified MPDs provide clients with an option to download parts of segments (chunks) from Web nodes (peers) which have the segments cached. However, unlike pDASH, DAV considers peer hosts inside a campus network where uplink characteristics need not be considered and consequently segments need not be “chunked”. Additionally, the utilisation of local content requires minimal firewall modifications (a local system administrator simply opens port 80 on client machines). Furthermore, while pDASH randomly selects peer hosts, DAV selects the best performing hosts for inclusion in the modified MPD, based on host rating. Apart from simplifying the decision-making process at the client, limiting the number of alternative hosts listed per segment also reduces the size of the
MPD file. Furthermore, the pDASH player requires an algorithm for concurrent download of peer-chunks and segments from servers, while DAV’s modified MPDs can be used with standard DASH players. pDASH focuses on reducing bandwidth utilisation, and client side evaluation results were not presented in [117]. Apart from modifying the original MPD at request time, the DAV Gateway provides dynamic MPD generation at each segment request. The latter approach outperforms the MPD modifications proposed in pDASH.
QDASH [122] utilises a hardware proxy hosting QDASH-abw [122] which accurately measures available link capacity to achieve gradual quality changes. While a QDASH-enabled video player maintains a “light-weight flow” [122] with the proxy to receive current measurements, the proxy does not provide further guidance in terms of hosting server selection. Furthermore, QDASH does not take the locality of the segments into account.
Similar to our solution, clients in NOVA [126] contact the network controller (centralised unit) to indicate segment download completion. However, NOVA clients do not provide information about locally stored content, so such content cannot be used by other clients.
Control Plane framework [136] also receives client side information, where active clients periodically (every few seconds) report quality statistics (e.g., buffering, join time, average bitrate) to the Framework’s Measurement Engine. However, the downloaded content is not utilised by other active clients.
6.1.2 Personalised Learning Systems
Online distributed systems, despite continuous hardware and network capacity improvements, remain vulnerable to delays, especially in settings where a high number of Web users access real time media. Open and distributed PL systems suffer from the same problem. Chapter 3 of this thesis presented a review of adaptive PL systems. One of the first families of well-defined and formally evaluated personalised online systems in the educational setting was Adaptive Educational Hypermedia (AEH) systems. These systems were investigated in Chapter 3 with a focus on their structure and adaptation approaches. AEH systems adapt learning material (in terms of content selection and presentation) to learner characteristics and learning context. Therefore, approaches to context-aware adaptation were outlined and PL systems supporting network and user device adaptation were explored.
Early AEH systems were not modularised and offered limited opportunity for improvement since, in most cases, modification and/or extension required full access to the system source. Network-awareness could be implemented by changing the system’s Presentation Model (PM) and Adaptive Engine (AE), so that the system considers network factors and user device. Third generation, service oriented systems, such as APeLS [197] addressed this issue, providing extensibility via new modules, such as the performance-aware solutions developed in this work. While AEH systems are online systems, potentially using distributed content, very few consider network/device characteristics when adaptation is performed and would benefit from the
solutions developed in this research. Solutions that perform adaptation based on user device and the underlying network conditions are limited to static content (e.g. QoE-aware AHA system (QoEAHA) [63]) or focus on device characteristics only (e.g. Mobile Mathematics Tutoring (MoMT) [240]). However, neither of these solutions considers transmission of video content. Solutions that deal with video [228] make adaptation decisions at the provider’s side. Our solutions focus on the learner side, which is an approach that scales better.