ABSTRACT
MPEG DASH provides formats that are suitable to stream media content over HTTP. Typically, the DASH client adaptively requests small chunks of media based on the available bandwidth and other resources.
This client-pull technology has proven to be more flexible, firewall-friendly and CDN- scalable than server-push technologies.
However, service providers have less control given the decentralized and client-driven nature of DASH, which introduces new challenges for them to offer a consistent and possibly higher quality of service for premium users.
This issue is addressed by MPEG in a new work referred to as SAND: Server and Network-assisted DASH.
The key features of SAND are asynchronous network-to-client and network-to-network communication, and the exchange of quality-related assisting information in such a way that it does not delay or interfere with the delivery of the streaming media content.
INTRODUCTION
Over the last few years, HTTP-based adaptive streaming has become the technology of choice for streaming media content over the Internet.
In 2012, MPEG published a standard on Dynamic Adaptive Streaming over HTTP (DASH) [1], which has been adopted and profiled by other standards and industry bodies, including DVB, 3GPP, HbbTV and DASH- IF.
The DASH formats are primarily designed to be used in client-pull based deployments with HTTP the protocol of choice for media delivery.
A client first retrieves a manifest in a Media Presentation Description (MPD), and then it selects, retrieves and renders content segments based on that metadata, as seen in Figure 1.
DASH when deployed over HTTP offers some fundamental benefits over other streaming technologies. DASH requests and responses pass firewalls without any problem, like any other HTTP messages.
As the content is typically hosted on plain vanilla HTTP servers and no specific media servers are necessary, DASH is highly scalable: DASH segments can be cached in HTTP caches and delivered via Content Delivery Networks (CDN), like any other HTTP content.
Most importantly, a DASH client constantly measures the available bandwidth, monitors various resources and dynamically selects the next segment based on that information.
If there is a reduction in bandwidth, the DASH clients selects segments of lower quality and size, such that a buffer underrun is prevented and the end user retains a continuous media consumption experience.
From many studies, it is well known that start-up delays and buffer underruns are among the most severe quality issues in Internet video and DASH constitutes a solution to overcome and minimize such problems.
However, the fundamental decentralised and client-driven nature of DASH also has some drawbacks.
Service providers may not necessarily have control over the client behaviour. Consequently, it may be difficult to offer a consistent or a premium quality of service.
Examples include that the resources announced in the MPD may become outdated after a network failure or reconfiguration, resulting in misdirected an unsuccessful DASH segment requests by the client.
A DASH client can mistakenly switch to lower quality segments, when a mobile hand-over or a cache miss is interpreted as a bandwidth reduction.
Massive live DASH streaming may lead to cascades of cache misses in CDNs. A DASH client may unnecessarily start a stream with lower quality segments, and only ramp up after it has obtained bandwidth information based on a number of initial segments.
Multiple DASH clients may compete for the same bandwidth, leading to unwanted mutual interactions and possibly oscillations [6].
As a consequence, service providers may not be able to guarantee a premium quality of service with DASH, even in managed networks where regular DASH clients may not fully take advantage of the offered quality of service features.
In 2013, MPEG and the IETF organised a joint workshop [2] to discuss the issues and potential solution directions, as input to the 2nd edition of the DASH standards.
Soon after, MPEG started the Core Experiment on Server and Network-assisted DASH (CE-SAND), in which use cases are defined and solutions are explored.
Based on the results of CE- SAND, MPEG is developing an architecture, data models and a protocol solution, expected to published as part 5 of the MPEG DASH standard in 2016. The use cases and status of the work as of mid of 2015 are summarized in the remainder of this paper.
No comments yet