Optimization of Video Serv rvices by SDN-Assisted Edge Computing - - PowerPoint PPT Presentation

Optimization of Video Serv rvices by SDN-Assisted Edge Computing

• A. Murat Tekalp

Department of Electrical and Electronics Engineering, Koç University, Istanbul, Turkey

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Contents

• Introduction
• Video Services over IP
• Edge Computing @NSP Edges
• Virtualization of Video Services @NSP Edges
• vCDN: SDN-Assisted HAS Services
• IPTV Services by SDN-Assisted IP Multicasting
• vSFU: SDN-Assisted WebRTC Services
• Managing Video Services @NSP Edges
• Emerging NSP-Managed Video Services
• Traffic Engineering at SDN-enabled NSP Edge
• Path Computation
• QoS Provisioning
• NSP-Managed HAS and WebRTC Services
• Conclusions
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Introduction – Evolution of IP Video

• CDN
• Akamai (1999), Google, NetFlix
• Improvements in Video Coding/Streaming
• MPEG2 (1996) MPEG4-AVC/H.264 (2003) HEVC/H.265 (2013) VP9/AV1
• RTP/UDP/IP push HTTP Adaptive Streaming - HTTP/TCP/IP pull (2007)
• Improvements in Networking
• Better Connectivity (wireline and wireless) 3G 4G 5G
• NSP Edge Computing
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

State of the Art in Video Services

• Over-The-Top (OTT)
• Offered by third party content providers over the open unmanaged Internet
• Providers rely on the best-effort service offered by the network service providers
• VOD, Live streaming, Interactive communications services
• IPTV
• Offered by network operators over managed IP networks
• Packets travel over closed IP network, IP multicast via the Central Office (PoP)
• Quality-of-Service (QoS) enabled differentiated video service
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Brief History of Edge Computing

• W. Shi, G. Pallis, Z. Xu, Edge Computing,

Proceedings of the IEEE, vol. 107, no. 8,

• pp. 1474-1481, August 2019.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Where is the Edge and Who Owns it?

• Operated by Cloud Service Provider – CDN: OTT video @ the network edge
• Best-Effort from the CDN node to end-user – No end-to-end quality of service (QoS)
• Operated by Network Service Provider – IPTV multicast @ NSP point of presence
• Can provide QoS from the edge to end-user over reserved network slices
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

• A. Vakali and G. Pallis, Content delivery networks: Status and trends, IEEE Internet Comput., vol. 7, no. 6, pp. 68-74, Nov. 2003.

IPTV: IP-Multicasting

• Multicasting is fundamental to the implementation of IPTV.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco Broadcasting

Network NOT multi-cast enabled

Three different channels IGMP query ON IGMP snooping ON

Multi-cast enabled network

IGMP snooping ON IGMP snooping ON Three different channels

IGMP- Internet Group Management Protocol

OTT Video Services

Network Service Providers OTT Video Service Providers End Users

Problem: Provide lower delay higher throughput service Problem: Receive better video quality and lower latency Video on Demand (HAS) Live Streaming (HAS) Real-Time Communications (WebRTC) Problem: Optimize network resource consumption, increase revenue Cloud or CDN

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

CDN: Physical Servers

Physical servers at centralized facilities and peering locations

• A. Murat Tekalp
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

OTT Video CDN @NSP Edge

Open Connect by Netflix

• Netflix partners with ISPs to localize substantial amounts of traffic by

deploying embedded Open Connect Appliances at edge points of presence.

• Netflix uses a popularity-based algorithm to determine what content should

be on the appliances. A number of factors affect popularity, and the contents

• f an appliance are expected to change on a daily basis.

OCA that is embedded in a partner network Peering without deploying embedded OCAs in the partner network

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

vCDN

Virtual CDNs leverage shared physical servers Virtual CDNs can be orchestrated – VNF and MANO-like orchestration

• A. Murat Tekalp

https://www.akamai.com/us/en/resources/library.jsp?type=t18&bm=a#stp=1

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Adaptive Video Streaming

• Adapt the video resolution to display capability of receivers
• Different spatial and temporal resolutions to different receivers,

e.g., 1080p, 720p, 480p, 360p

• Adapt the video bitrate to available network rate
• Different bitrates to different receivers
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Best-Effort Streaming Services: Serve at multiple resolutions and bitrates

Video Coding for Adaptive Streaming

Example: Two spatial and three temporal layers

• Stream Switching
• Encode source video at multiple resolutions and bitrates

e.g., 1080p, 720p, 480p, 360p

• Scalable Video Coding (SVC)
• Multiple spatial and temporal resolution layers in the same video bit stream
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

HTTP Adaptive Streaming (HAS)

• When a client requests a video title, the server first sends an MPD file,

which lists available video chunks encoded at different resolutions and rates.

• The client app requests video chunks whose resolution and bitrate match

its display resolution and current available channel rate.

• Problems:
• Server-side congestion: Video service providers deploy or rent CDN clouds

to avoid server congestion and provide lower service latency.

• Network congestion: Multiple HAS flows on the same link compete with

each other for available bitrate. TCP congestion control mechanism causes unwanted rate/ video quality fluctuations.

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Common Media Application Format(CMAF)

• A. Murat Tekalp

Compatible with AV1 video codec Alliance for Open Media (AOM)

• Unified packetizer format based on fragmented MP4 (fMP4)
• Apple HLS used to use M2TS packetization. With CMAF, the use of M2TS for video streaming is deprecated.
• The codec and DRM wars still continue …
• Non-multiplexed and precisely

aligned (synchronized) streams

• Encryption (CENC: CBC or CTR)
• Audio codecs (AAC or optionally

multi-channel)

• Captions, subtitles (WebVTT or ISMC1)

https://www.unified-streaming.com/blog/promises-cmaf-and-its-compatibility-unified-streaming-platform

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Real-Time Communications (RTC)

• Media streams
• P2P between two peers
• Multi-party RTC service architectures
• Session initiation
• Signaling Server
• Problems:
• Guaranteed low-delay service
• Efficient use of network resources

Mesh Non-scalable (default) or scalable video MCU Nonscalable video Transcoding at the MCU SFU Scalable video Cloud

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Mesh-Connected WebRTC Services

• Default: Clients run non-scalable video codec
• Adaptive: Clients run a SVC enabled encoder
• Ability to adapt different receiver terminal types.
• Ability to adapt different receiving bandwidth capacities.
• G. Bakar, R. A. Kirmizioglu, and A. M. Tekalp, “Motion-based rate adaptation in WebRTC videoconferencing using

scalable video coding,” IEEE Trans. on Multimedia, vol. 21, no. 2, pp. 429-441, Feb. 2019.

• Advantages
• Easy to implement.
• Low end-to-end latency.
• Drawbacks
• Limited peer upload capacity
• Limiting receiving peer factor (in case of non-scalable video coding)

No Cloud or Edge Server

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Best Effort

SFU-Connected WebRTC Services

• Advantages
• Single video upstream architecture (overcomes limited peer upload capacity problem)
• SVC enabled service for different receiver terminal types.
• Drawbacks
• High end-to-end service latency.
• High network resource consumption.
• Single point of failure.
• Traffic overload since all video traffic must go through the SFU.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Best Effort Selective Forwarding Unit Media Server

MEC

• Mobile Edge Computing
• End users off-load computation from mobile devices to edge servers
• Multi-access Edge Computing
• Telcos use edge servers to manage fixed and radio access networks
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

• A. Wang, Z. Zha, Y. Guo and S. Chen, Software-defined networking enhanced edge computing: A network-centric survey,

Special Issue of the Proc. of the IEEE on Edge Computing, vol. 107, no. 8, pp. 1500-1519, August 2019. Multiple access capability: Users can access the same service deployed at the edge of the network and obtain the same user perception no matter whether they access the network through 4G, 5G, or even fixed networks. Compute capability: Edge servers and accelerated hardware devices such as GPU/FPGA, together with the edge ICT cloud platform, provide computing, storage, and acceleration capabilities to applications at the edge. Service enablement capability: Enables provisioning carrier-grade or third-party services at the edge.

Network Technology Enablers

• Software Defined Networks (SDN)
• Cloud service providers and datacenters employ SDN
• Network Function Virtualization
• Network service providers employ SDN and NFV
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

• Services at edges can be configured as

virtual network functions (VNF) that can be centrally orchestrated.

• Services at edges are mainly third-party
• services. Manual network configuration for

services requires huge O&M workload.

• The unattended operation of some edge sites

makes network configuration more difficult.

• SDN enables programmable automatic

network configuration and traffic engineering at edges.

Network Industry Trends

• Cloud service providers are moving towards network edges

for lower latency and higher bandwidth access to end users (also known as fog computing)

• Network service providers (NSP) are replacing closed and

proprietary hardware-based access technologies with disaggregated and virtualized software running on edge clouds to manage their edge access networks.

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

CORD Architecture

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco Software architecture

Open network Foundation (ONF) Project

• Leverages open source projects such as OpenStack, ONOS, XOS

Central Office Re-architectured as a Datacenter

SDN-Enabled Multi-Access Edge

https://www.opennetworking.org/seba/

• SEBA is an exemplar lightweight platform built

by the ONF and CORD community that supports a multitude of virtualized access technologies at the edge of the carrier network, including PON, G.Fast, DOCSIS and more.

• SEBA supports both residential access and

wireless backhaul and is optimized such that traffic can run ‘fastpath’ straight through to the backbone without requiring VNF processing on a server.

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Multi-Access Edge

• A. Murat Tekalp

Virtual OLT support in SEBA – vendor independent ONT at customer premises

vCO Architecture

Virtual Central Office

Linux Foundation Project

• Based on ETSI standard
• vCO 1.0
• Residential Services and Enterprise

Business Services live on stage (vOLT, VNFaaS, BNG, etc)

• Generic blueprint for Central Office

with open source components and OpenDaylight SDN controller

• vCO 2.0
• Mobile Services
• A. Murat Tekalp

NFV Orchestrator, Service Assurance (Met Mon) Element Manager VNF(s) vBNG NFVI VNF Manager

ONAP Orchestrator

SDN Controller White Box Switch Fabric Servers/HW VNF(s) vRouter VNF(s) FW VNF(s) FW Element Manager Element Manager

https://www.opnfv.org/resources/virtual-central-office

Proprietary Solution: ZTE COMMON EDGE SOLUTION ARCHITECTURE

ZTE-MEC White-Paper - Published Oct. 2019 https://sdnfv.zte.com.cn/en/insights/2019/10/ZTE-Common-Edge-White-Paper https://sdnfv.zte.com.cn/download/ZTE%20Common%20Edge%20White%20Paper.pdf

• A. Murat Tekalp

Why Video Services @NSP Edge?

• Video Service Providers
• Lower end-to-end delay
• Ability to provide premium service with lower delay and higher

bandwidth.

• Network Service Providers:
• More efficient use of its network resources.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

IPTV service OTT: HAS service OTT: WebRTC service

vCDN @ NSP EDGE

Can OTT video services match the quality of IPTV ? You should use the same weapons as your competitors

• A. Murat Tekalp
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

IPTV @NSP Edge

• All video streams are delivered up to the NSP edge server
• IP-multicast groups and IP-multicast stream routing are

managed at the compute nodes of the NSP edge server.

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

RTC SFU @NSP Edge

• Provides advantages of mesh-connected architecture while
• vercoming limited peer upload capacity problem
• Allows scalable coding without overloding peers with

intelligence required to manage other peer video capability. Decisions on who receives which layers moves to the edge.

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Place SFU at network edges

Emerging Managed Video Services

• Network Service Providers:
• OTT providers are becoming overwhelming loads on the network operators. Video

streaming over the IP traffic is increasing, but not directly contributing to the revenue of the telecom operators.

• Main revenue stream of the telecom operators are clients paying flat subscription fees in

exchange for the best-effort Internet access. Light users subsidize heavy users when total revenue considered.

• Leverage SDN infrastructure to manage efficient use of network resources and collect

more revenue.

• Video Service Providers:
• Ability to provide premium service with lower delay and higher bandwidth.
• End Users:
• Choice of receiving higher quality of service for an additional fee.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Why Manage at the Edge PoP?

• High-bandwidth connection between main video server and CDN servers.
• Main sources of congestion
• Video server (HAS) or SFU (RTC)
• Access network
• Solution:
• Video CDN or SFU as Edge-native Applications
• Leverage SDN infrastructure for virtual network slicing and QoS
• Manage Premium Video delivery over Special Network Slices with

guaranteed low-delay and/or guaranteed bandwidth

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Traffic Engineering Module (TEM)

VSP APP SDN Controller TEM

• It is implemented as a Web Service.
• It communicates with other components using

the REST interface.

• Input
• link/queue state informations
• QoS request informations Constrained shortest

path algorithm to determine the routes between the peers and CDN node according to scheduling decisions.

• Output
• Calculated paths (sequence of switch queues)
• Group reservation for group of peers to the CDN

node, which will guarantee QoE requests between the CDN node and the peers.

• Reservation per client is not efficient.

33

‘Scheduling decisions’ ‘Scheduling decisions’ ‘Topology information’

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

QoS Provisioning (Queue Allocation)

• SDN switches support queues providing per-flow rate-limiting

ability; hence, make QoS implementations possible.

• Switch ports are configured with multiple queues at different

packet egress rates using OVS commands.

• Packets of a particular traffic is assigned to an appropriate

queue through OpenFlow flow modification messages.

• Queue Management
• Fixed number of queues at fixed capacities (suitable for hardware

switches)

• Dynamic number of queues at variable capacities (suitable for software

switches)

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

NSP-Managed HAS Streaming Services

35

Edge Access Network

SDN Controller

Aggregation Switch

Multi-Access Edge Computing Unit

Peer Peer Peer

Core Network

VSP Embedded in NSP-Edge

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

SDN-assisted delivery service

NSP Edge Access Network

NSP-Managed WebRTC Services

• Session Initiation
• Constrained shortest path computation.
• Bandwidth reservation.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

• Guaranteed bandwidth to all parties.
• Minimum end-to-end delay between all pairs of parties.
• Minimum overall network bandwidth consumption.

SFU-Connected WebRTC Services by SDN-assisted Edge Computing

Cloud SFU Service

• All parties are connected by a single SFU
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Edge SFU Service

• Distributed multiple SFUs in a single session.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Simulation Environment

• NSP Simulation – Random Topology
• Core network: 256, 400, 576, 784, 1024, 1296, 1600 switches.
• Edge networks: 16, 25, 36, 49, 64, 81, 100 switches.
• Distance based probabilistic model:
• WebRTC Service Simulation
• Maximum E2E delay of all WebRTC sessions
• The total bandwidth consumption of WebRTC services over the edge and core networks
• Mean of 100 trials.
• Sessions with 3, 4 and 5 peers with total of 100 sessions.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco (256c/16e)

Performance Evaluation

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Conclusions

• Cloud-based OTT services are subject to delay and network unstability.
• Next generation video services will be delivered from SDN-enabled

multi-access edge points of presence.

• OTT video service providers will collaborate with Network service

providers to offer premium managed services much-like IPTV services.

• Content distribution networks at multi-access edge points
• Interactive communications services servers (SFU) at multi-access edge points
• Video services over reserved edge-network slices
• New technology will enable new video service business models.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Publications

• R.A. Kirmizioglu and A.M. Tekalp, "Managed multi-party WebRTC videoconferencing services by SDN-assisted IP multicasting of scalable encoded video", IEEE Trans. on

Multimedia, accepted for publication.(available as pre-publication)

• R.A. Kirmizioglu, A.M. Tekalp, "Multi-party WebRTC as a managed service over multi-operator SDN" (will be presented on Friday)
• R.A. Kirmizioglu, A.M. Tekalp, B. Gorkemli, "Optimization of Multi-Party WebRTC Videoconferencing Service by SDN-assisted Edge Computing", IEEE Journal on Selected Areas in

Communication (JSAC) (in review).

• K.T. BagciandA. M. Tekalp,“SDN-enabled distributed open exchange: Dynamic QoS-path optimization in multi-operator services,” Elsevier Computer Networks, vol. 162, Oct. 2019.
• G. Bakar, R.A. Kirmizioglu, A.M. Tekalp, "Motion-Adaptive Rate Control in WebRTC Videoconferencing using Scalable Video Coding", IEEE Transactions on Multimedia, vol. 21,
• no. 2, pp. 429-441, Feb. 2019.
• R.A. Kirmizioglu, B.C. Kaya, A.M Tekalp, "Multi-Party WebRTC Videoconferencing Using Sacalable VP9 Video: From Best-Effort Over-The-Top to Managed Value-Added Services",

IEEE Int. Conf. on Multimedia and Expo (ICME), Jul. 2018.

• G. Bakar, R.A. Kirmizioglu, A.M. Tekalp, "Motion-based Adaptive Streaming in WebRTC using Spatio-Temporal Scalable VP9 Video Coding", IEEE Global Communication Conference

(Globecom), Dec. 2017.

• K. E. Sahin, K. T. Bagci, and A. M. Tekalp, "Distributed-collaborative managed DASH video services," Int. Conf. Network and Service Management (CNSM), Tokyo, Japan, Nov. 2017.
• K.T. Bagci, K.E. Sahin, and A. M. Tekalp, “Compete or collaborate: Architectures for collaborative DASH video over future networks,” IEEE Trans. on Multimedia, vol. 19, no. 10,
• pp. 2152-2165, October 2017.
• K. T. Bagci, S. Yilmaz, K. E. Sahin, A. M. Tekalp, “Dynamic End-to-End Service-Level Negotiation over Multi-Domain Software Defined Networks,” IEEE Int. Conf. on Communications

and Electronics (ICCE), Ha Long, Vietnam, 2016.

• K. T. Bagci and A. M. Tekalp, “Managed Video Services over Multi-Domain Software Defined Networks,” European Signal Processing Conf. (EUSIPCO), Budapest, Hungary, 2016.
• K. T. Bagci, K. E. Sahin, A. M. Tekalp, “Queue-Allocation Optimization for Adaptive Video Streaming over Software Defined Networks with Multiple Service-Levels,” IEEE Int. Conf.
• n Image Processing (ICIP), Phoenix, USA, 2016.
• S. Civanlar, E. Lokman, B. Kaytaz, and A. M. Tekalp, ``Distributed management of service-enabled flow-paths across multiple SDN domains," European Conf. on Networks and

Communications (EuCNC), Paris, Fransa, June 2015.

• S. Yilmaz, A. M. Tekalp, and B. D. Unluturk, ``Video streaming over software-defined networks with server load balancing," Int. Conf. on Computing, Networking and

Communications (ICNC), Anaheim, CA, USA, February 2015.

• H. Egilmez and A. M. Tekalp, "Distributed QoS architectures for multimedia streaming over software defined Networks," IEEE Trans. on Multimedia, vol. 16, no. 6, pp. 1597-1609,

October 2014.

• H. Eğilmez, S. Civanlar, and A. M. Tekalp, “An optimization framework for QoS-enabled adaptive video streaming over OpenFlow networks,” IEEE Trans. on Multimedia, vol. 15,
• no. 3, pp. 710-715, April 2013.
• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco

Questions?

• A. Murat Tekalp

November 20-22, 2019 ACOSIS 2019 Marrakech, Morocco