SDNDASH Improving Qo E of HTTP Adaptive Streaming

SDNDASH: Improving Qo. E of HTTP Adaptive Streaming Using Software Defined Networking Presented by Tuo Yu 09/28/2017 1

Motivation 2

Motivation • Dynamic Adaptive Streaming over HTTP (DASH): • A video on a server is segmented into chunks (2 -10 seconds). • Each chunk is encoded at several different bitrate levels and resolutions. • Clients fetch chunks from servers. • Each client uses its bitrate adaptation logic to dynamically fetch the chunk encoded at the optimal bitrate level based on metrics such as the average throughput and the buffer occupancy. 3

Motivation • Currently DASH is sub-optimal: • Inaccurate available bandwidth estimations result in frequent quality and bitrate oscillations. • Clients compete for the bandwidth, and thus service differentiations (e. g. , gold, silver and bronze clients) and management policies cannot be assured on the delivered quality. 4

Challenge and Approach • A robust DASH scheme should assure: quality stability, fairness and high utilization. • SDNDASH • It is based on the centralized management. • It leverages SDN to select for each client the optimal quality and bitrate level for the next chunks. • It maximizes the per-client Qo. E (Quality of Experience) in a fair manner, and improves the per-client network resource allocation, provisioning and monitoring. 5

Related Work • Using the Buffer to Avoid Rebuffers: Evidence from a Large Video Streaming Service (T. -Y. Huang, et al. ) • A buffer-based rate selection algorithm (BBA) • The bitrate selection heuristic is based on the playback buffer occupancy instead of the bandwidth. • It maximizes the average video quality and avoids unnecessary rebuffering events. • However, this scheme decreases the Qo. E in the case of longterm bandwidth variations and slow bandwidth fluctuation detection. 6

Related Work • QDASH: A Qo. E-aware DASH system (R. K. P. Mok, et al. ) • A Qo. E-aware system as a proxy between the clients and the streaming server (QDASH) • It avoids video quality oscillations by ensuring a gradual change in bitrate levels. • It adds an intermediate level into the bitrate switching process instead of suddenly switching up/down to the target bitrate level. • However, QDASH generates significant network overhead if the number of clients is large. 7

Architecture 8

DASH Server • DASH Server: an HTTP server that stores the media chunks. • Chunk quality file: it lists chunks with their perceptual quality, which is measured based on the SSIMPlus index objective metric. 9

DASH Clients • DASH Client: • It receives the recommended optimal bitrate level, quality and buffer level from the external SDN-based resource management application. • Its bitrate adaptation logic uses these values as the upper bounds. 10

External SDN-based Resource Management • External SDN-based Resource Management Application: • Performs Qo. E-optimization, resource allocation and monitoring. 11

Qo. E Metrics • Startup Delay ( ): the delay between the request of the first chunk and its rendering onset. • Number of Stalls ( ): the number of stalls during the play of the video. Stalls occur when the client buffer is empty. • Average Video Quality ( ): the total average quality of the downloaded chunks. The quality of a chunk is evaluate by SSIMPlus. • Video Quality Switches ( ): the total quality differences between successively downloaded chunks. Constants • Qo. E is the weighted sum: 12

External SDN-based Resource Management • Measures the Qo. E of each DASH client and estimates the available bandwidth. • Maximizes per-client Qo. E and outputs the optimal per-client bitrate level and quality for the next chunks to be downloaded. 13

Objective Function • At each step i, and for each client , calculate: l: bitrate q(. ): corresponding quality, specified by SSIMPlus Maximize the Qo. E of the client 14

Objective Function The selected bitrate must not exceed the current available bandwidth. The sum of the allocated bandwidth not exceed the total available bandwidth. The selected bitrate level and quality must not be affecting the buffer occupancy. 15

Objective Function The selected chunk quality must fit the content type (e. g. , animation, documentary). The decision of bitrate level and quality must fit the device capabilities. • The device must be able to play the chunks with the selected bitrate level and quality. 16

Objective Function • It can be solved using online MPC (Model Predictive Control). • The total complexity is independent of the total number of clients. 17

External SDN-based Resource Management • Measures the Qo. E of each DASH client and estimate the available bandwidth. • Maximizes per-client Qo. E and outputs the optimal per-client bitrate level and quality for the next chunks to be downloaded. • Formulates the Qo. S policy based on the optimal per-client Qo. E. Video codec parameter. Network layer parameters defined based on the Qo. S service differentiation (gold, silver and bronze clients). Typo: are reused in the paper. 18

External SDN-based Resource Optimal bitrate level, quality and Management buffer level for each client Qo. S policy for each client 19

Internal SDN-based Resource Management • RYU SDN Controller & Internal Application • It is a central entity that manages flow control. • It receives the per-client Qo. S policy from the external SDN-based application. • It constructs an Open. Flow based messages for each Qo. S policy, and sends them to the forwarding devices. 20

Forwarding Devices • Forwarding Devices (switches, routers) • They forward data traffic based on the received Open. Flow messages. 21

Evaluation • 50 clients and 100 Mbps total network bandwidth. • Baselines: QDASH, BBA, SARA, and conventional bitrate adaptation heuristic of dash. js 22

Evaluation • Video Stability 23

Evaluation Steps • Fairness N-Qo. E: Normalized Qo. E 24

Evaluation • Qo. E 25

Evaluation • Utilization 26

Conclusion • SDNDASH maximizes per-client Qo. E, whilst taking into account different clients' needs (e. g. , buffer sizes, quality requirements) and network requirements (e. g. , available bandwidth). • The experimental results show that SDNDASH provides a good network stability with optimized end-user Qo. E across all the clients in a shared network environment. 27

Pros and Cons • Pros: • SDNDASH is the first work that propose a complete end-to-end architecture to address DASH scalability in the context of a shared network. • The capability of devices is also considered during the design of SDNDASH. • • Cons: The cost of the SDN-based centralized management is not considered. The content type constraint is not clearly introduced. 28 The design of SDNDASH supports heterogeneous clients (e. g. ,

Pros and Cons • Is it fair to include the settling time of the baselines? 29

Thank you 30

Motivation • Dynamic Adaptive Streaming over HTTP (DASH): Thomas Stockhammer, Qualcomm, “DASH –Design Principles and Standards, Presentation at MMSys 2011 31

External SDN-based Resource Management • Measures the Qo. E of each DASH client and estimate the available bandwidth. • Maximizes per-client Qo. E and outputs the optimal per-client bitrate level and quality for the next chunks to be downloaded. • Formulates the Qo. S policy based on the optimal per-client Qo. E. Video codec parameter. Network layer parameters defined based on the Qo. S service differentiation (gold, sliver and bronze clients). Typo: are reused in the paper. 32