5g grand research challenges · • nfv-based network softwarisation/slicing • provisioning of...

16
5G Grand Research Challenges Friday, 17 November 2017 1 Ning Wang 5G Innovation Centre

Upload: others

Post on 31-Dec-2020

2 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

5G Grand Research Challenges

Friday, 17 November 2017 1

Ning Wang

5G Innovation Centre

Page 2: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

5G is…

2

CONNECTIVITY + INTELLIGENCE

Data to information/knowledge transformation

Blurring boundaries between real and cyber worlds

Connected Devices of small and large sizes and capabilities(robots, cars, sensors, actuators, smart phones ………. driverless cars)

Automation

Page 3: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

5G – A Special Generation

3

• Previous “G”s: different versions of dummy bit pipe with

evolved bandwidth capacities in transmitting data

• Key features of 5G networks

Network slicing: simultaneously support a wide range of

services with specific requirements

Enhanced mobile broadband

Ultra-reliability and low latency

Massive connectivity

Softwarisation and Virtualisation

Network resources

Network functions

Edge computing

Page 4: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

Computing vs. Communication in 5G

4

Computing

5G is a complex ecosystem with cooperative computing and

communication operations

Communication

OTT 5G Applications & Services

Service assurancesupport

Network resourceefficiency

5G network resources (radio, bandwidth, storage, CPU…)

Computing for Communication: In-time computing (e.g. signal processing) to boost data transmission rate

Communication for computing: In-time delivery of data across distributed computing elements (edge computing) to be processed

Page 5: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

Context-aware Network Operations in 5G

5

5G Network Architecture

Virtual network functions (VNFs)

Contextdriving

Contextharvesting

User context

Device context

Network context

Content context

• Networking protocol design with enhanced or bounded performance (under given conditions)

• Addressing fundamental limitations of existing TCP/IP paradigm

• NFV-based network softwarisation/slicing

• Provisioning of virtual network functions (VNFs) for handling different requirements

• Context data harvestingand profiling

• Lightweight online userQoE inference

• Building knowledge to control network functions…

Tuesday, 19 September 2017

Virtual network functions (VNFs)

Statistics on:

Page 6: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

6

• Network traffic patterns

• User mobility patterns

• Mobile content consumption patterns

Examples on Statistics

Page 7: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

7

Learning Traffic Patterns (Network load)

0

0.2

0.4

0.6

0.8

1

1.2

1 101 201 301 401 501 601

Max

imum

Lin

k U

tiliz

atio

n

Interval

Profile 0

Profile 1

Peak-time weekday

Off-Peak-time Peak-time weekend

0

0.2

0.4

0.6

0.8

1

1.2

1 101 201 301 401 501 601

Max

imum

Lin

k U

tiliz

atio

n

Interval

Profile 0

Profile 1

Peak-time weekday

Off-Peak-time Peak-time weekend

(a)

(b)

• 7-day traffic load

dynamicity in a

typical operational

network (Not 5G but

still reflects how

users use the

network on daily

basis!)

• Knowledge can be

derived from traffic

statistics for

optimising network

efficiency

• Use case – putting

network elements to

sleep mode for

energy efficiency

Page 8: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

8

Learning Traffic Patterns (Application Distribution)

• Network traffic predictions – e.g. Cisco’s Visual Network Index (VNI) tool https://www.cisco.com/c/en/us/solutions/service-provider/visual-networking-index-vni/index.html

• 5G resource management for network slicing

Page 9: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

9

User Mobility Awareness and Prediction

• Reduction of signalling messages when a user enters a new tracking area

(TA) – up to 80% reduction depending on TA and TA List configurations

• Proactive content downloading against anticipated signal strength

deterioration for on the move content consumers

• Resource discovery in Device-to-device (D2D) communications

Page 10: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

10

• Motivation: To obtain accurate knowledge about content

consumption pattern both at the regional level and individual

level.

• Application scenarios:

• Content can be cached at the local data centres to be served to

a large number of users with common content interests

• Content can be preloaded at intelligent home gateways even

before the users start to consume

• Context-aware transmission mode switching: unicast

multicast upon detection of increased content popularity

• Research objective

• To design fast and accurate machine learning algorithms in

order to predict crowd/individual-based content consumption

behaviours

Content Profiling and Prediction

Page 11: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

11

• Scheme overview

• Continuous window-based training according to historical data and

make predictions on near future events

• Key factors to be considered: similarity between users in content

consumption, content popularity dynamicity, correlation between

different dimensions of dynamicity on the content side and the user

side

• Technique under patent consideration

• Key results

• The average of RMSE for all popularity predictions is 0.047

(Normalized by total number of actual number of requests)

• Relative improvement of more than 50% in comparison with three

well-known SoA solutions (Szabo-Huberman, Multivariate Linear

Model and its extension)

Content Profiling and Popularity Prediction

Page 12: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

12

• Content-level popularity: heavy-tail – top 10% popular content attracted

90+% requests

• The majority of today’s video applications use DASH (Dynamic Adaptive

Streaming over HTTP) in which the whole content is divided into fixed-length

chunks/segments

• Caching Video-on-Demand (VoD) video segments at the 5G mobile edge

Content Popularity and Content Chunk Popularity

Page 13: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

13

Towards Emerging Augmented / Virtual Reality Applications

From 3 Degree of Freedom (3DoF) to 6DoF

User behaviour patterns in AR/VR applications are much more complex than traditional media – generation of explosive context data on user behaviours

Public

Internet

VR content

cloud

5G mobile

edge

5G mobile

edge

Live streaming content

Embedded 5G network functions

• Prefetching

• Caching

• Transcoding

• Rendering

• User behaviour capturing

• Content popularity analysis

• …

Assuring user experiences

• Low start-up delay

• Guaranteed content

quality (resolution)

• No playback stalling

• Minimum streaming

latency

• …

Page 14: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

14

Capturing User Behaviours in Immersive Applications

• Selective transmission of content

• In some 360-degree video the whole picture is partitioned into multiple “tiles”

• Only the tiles covering the actual Field of View (FoV) is useful

• Applicable content manipulation techniques

• Caching: cache popular tiles at the mobile edge based on crowd-sourced

knowledge

• Prefetching: Online prediction of the use behaviour and prefetch follow-up tiles

that are most likely to cover the FoV in the next moment

• Network intelligence

• Learning and prediction algorithms

• Enabling 5G technology

• Multi-access edge (Fog) computing

FoV

Page 15: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

15

• 5G – an ecosystem that can generate rich data

• Highlights on key research challenges

• Data harvesting mechanisms

• Multi-dimensional correlation of context-data produced by

different players in 5G

• Architectures for enabling “in-network” data analytics

• Hierarchical vs. peer-to-peer computing

• Offline/online operations

• Offline: long-term construction of network intelligence

• Online: real-time learning and instantaneous decision-making in

changing network behaviours – Possible in practice?

• Enabling network autonomics

Summary

Page 16: 5G Grand Research Challenges · • NFV-based network softwarisation/slicing • Provisioning of virtual network functions (VNFs) for handling different requirements • Context data

16

Thank You!