Slicing and Orchestration in Service-Oriented 5G Networks Navid Nikaein Associate Professor - Eurecom Coordinator of Mosaic-5G Initiative Keynotes at IEEE CAMAD 17-19 September, 2018, Barcelona, Spain. (c) Navid Nikaein 2017 1
2 Connected, Controlled, and Flexible Value Creation Consistent experience Sustainable business model 5G Promises (c) Navid Nikaein 2017 2
3 Source: Dohler Evolution to Internet of Skills (c) Navid Nikaein 2017 3
Communication-Oriented Servers Data-centers 4 Today s 4G is designed for a limited number of UCs - Throughput-optimized - Fixed - Rigid Nokia Is 4G enough? (c) Navid Nikaein 2017 4
Mindful about 3GPPP facts and figures 514 Companies from 45 Countries 50,000 delegate days per year 40,000 meeting documents per year 1,200 specifications per Release 10,000 change requests per year LTE 5 3GPP Communication-oriented 4G (c) Navid Nikaein 2017 5
Turn physical infrastructure into multiple logical networks, one per service instance: One-Network, Many-Service 6 NOT a one-size fits all architecture NOT a Dedicated Network Multi-Service Network Service-oriented 5G (c) Navid Nikaein 2017 6 Ericsson WP
Different aspects of network slicing have been already prototyped/demonstrated based on Opensource and commercials platforms 7 Industry is currently providing network slicing by means of (a) Local/dedicated services enabled by MEC platforms (b) Dedicated core networks and 3GPP RAN sharing Next steps : SO-CN and SO-RAN From R&D to Reality (c) Navid Nikaein 2017 7
Coherent Project 5G technology enablers (c) Navid Nikaein 2017 8 8
Network Slicing Concept 9 Flexible & Customizable logical networks tailored to each use-cases/verticals xmbb urllc mmtc Service-oriented 5G (c) Navid Nikaein 2017 9
Slicing Technology Enablers 10 - Softwarization - Virtualization - Disaggregation Multi-service multi-tenant network Service-oriented 5G (c) Navid Nikaein 2017 10
11 Why will it happen? Extreme network flexibility and elasticity Service-oriented 5G (c) Navid Nikaein 2017 11
Network Slicing evolves the value-chain of telecom industry Digital Service Provider (e.g. OTT, media, social, apps) Content-aware service optimization Communication Service Provider (e.g. operators, verticals) 12 Infrastructure Selection & Performance optimization Decoupling of Players, but the reality might be different Network Function Provider (e.g. vendors, verticals, 3 rd parties) Interoperability Network Infra. Provider (e.g. operator, vendor) Facility Availability & Compatibility Hardware Provider (e.g. IC designer, IDM) Hardware supportability Data Center Service Provider (e.g. operator, cloud, IT) Service-oriented 5G (c) Navid Nikaein 2017 12
13 3GPP Role Model (3GPPP 28.801) 3GPP Service-oriented 5G (c) Navid Nikaein 2017 13
3GPP Re-Architects Mobile Network 14 3GPP Service-oriented 5G (c) Navid Nikaein 2017 14
3GPP Re-Architects Mobile Network 15 3 Tier RAN Node CU0 DU[0-n] RRU[0-m] Functions Split CP - UP split Service-Oriented CN NEF Nnef NRF Nnrf PCF Npcf UDM Nudm AF Naf service catalog and discovery Slice selection function CP - UP split Nausf Namf Nsmf AUSF AMF SMF N1 N2 N4 UE (R)AN N3 UPF N6 DN 3GPP 5G RAN and CN (c) Navid Nikaein 2017 15
3GPP Re-Architects Mobile Network Slicing Functions 16 NSSF NEF NRF PCF UDM AF AUSF AMF SMF DU UEs UPF DN DU CU Control Plane User Plane AMF Access & Mobility Management Function SMF Session Management Function AUSF Authentication Server Function UPF User Plane Function NRF Network Repository Function AF Application Function UDM Unified Data Management PCF Policy Control Function NSSF Network slice selection function NEF Network Exposure Function 3GPP 5G RAN and CN (c) Navid Nikaein 2017 16
UDM NSSF NRF PCF 17 UPF3 AutoPilot SMF3 URLLC Slice UPF2 Caching AMF SMF2 embb Slice SMF1 UPF1 DN Default Slice miot Slice Maintenance/statistics miot, low throughput Infotainment/video streaming embb (Mobile Broadband) Safety/autonomous driving service URLLC (Ultra Reliable Low Latency) Dedicated or Shared Functions? (c) Navid Nikaein 2017 17
18 Network Slicing Dedicated or Shared? (c) Navid Nikaein 2017 18 M. Marina
Multiplexing Gain 19 Network slicing Dedicated or Shared? (c) Navid Nikaein 2017 19
20 Sharing Shared Resources? (c) Navid Nikaein 2017 20
Dedicated Resources? (c) Navid Nikaein 2017 21 21
Automation-Orchestration (c) Navid Nikaein 2017 22 22
23 LifeCycle Management (Encapsulate Operations) Interplay among control, management, and orchestration subsystems Automation-Orchestration (c) Navid Nikaein 2017 23
24 LifeCycle Management (Encapsulate operation) https://jujucharms.com/q/oai Automation-Orchestration (c) Navid Nikaein 2017 24
25 Phase Change of Modern Software Operation cost Software cost Free software is becoming expensive! Automation-Orchestration (c) Navid Nikaein 2017 25
Cloud & NFV Place holder for : RAN Optimization Network slicing Network model IoT Gateway Application MEC SDN C-APPS Content caching CN Control & Data Plane separation Abstraction & Programmability Application SDK RAN Control & Data Plane separation Abstraction & Virtualized Control Functions RESTful API 26 Network LTE enb OpenFlow switch Ethernet switch Network Abstractions and (c) Navid CP-Apps Nikaein 2017 Ecosystems 26
Data-driven Network Control (c) Navid Nikaein 2017 27
Network Slicing brings network flexibility and resource elasticity (1) Openup the interfaces with the help of SDN (2) Customized control Apps for monitoring, reconfigurability, and programmability 28 But, modern networks are too complex to be controlled and optimized by means of rulebased Alg. Why do we need to evolve 5G? (c) Navid Nikaein 2017 28
29 Flexibility to generalize and comprehend: Never seen Z before, but it is similar to X, so do Y, but adjust as needed Scale to automate control and management to meet the required QoS/QoE Dynamicity to constantly adapt and anticipate for different workloads and use cases Abstraction and multi-layering to combine sources with different semantics Why do we need to evolve 5G? (c) Navid Nikaein 2017 29
30 S. Katti What is Data-driven Control? (c) Navid Nikaein 2017 30
31 ML models to manage network and resources (a) Comply better with slice SLAs (b) Maximizes the revenue of physical network operators (c) Robust against runtime issues Pipelining by means of Reason-Predict-Control Why Data-driven Network Control? (c) Navid Nikaein 2017 31
32 Need to predict the user and network performances in time and space with many unknown and/or dynamic variables (1) Realtime control and coordination across cells (2) Network Intelligence Data-Driven Control (c) Navid Nikaein 2017 32
33 Smoother bitrate adjustment No buffer freezes Objective: maximize video quality minimize stall time Maintain service continuity Policy: maintain SLA (e.g. minimum average throughput) Data: Link quality sustainable TCP throughput Control (beyond just ABR, joint UE and BS): 1. Adapt the video bit rate through video optimizer 2.1 Add/provision a new BS through SMA+ORCH 2.2 increase the BW of the current BS (SMA) 3. Interference coordination through RRM 4. Update frequency and power through SMA Video Streaming Use-Case Revisited (c) Navid Nikaein 2017 33
Raw Info Knowledge Base (continuously updated) Policies Intents SLA 34 Monitor Future State Query Reason predict State Future Effect Control Proactive Control Scheme (c) Navid Nikaein 2017 34
Example of QoE With PCS (c) Navid Nikaein 2017 35 35
Example of QoE With PCS (c) Navid Nikaein 2017 36 36
Example of QoE With PCS (c) Navid Nikaein 2017 37 37
Example of QoE With PCS (c) Navid Nikaein 2017 38 38
Realtime Batch streaming Posts Exabytes Transactional Records, files 39 Velocity Volume Variety Veracity Structured Semi-structured Unstructured Availability Accountability Trustworthiness Characterizing the Data (c) Navid Nikaein 2017 39
Conclusion
Fusion of Computing, Information and Cellular technologies 41 (a) 5G and beyond is not only New Radio and verticals, it is also an evolution in General-Purpose computing for wireless networks (b) More and more software technologies (NFV,SDN,MEC) and Data (mining, analytics) jointly with radio signal processing Conclusion
Conclusion (c) Navid Nikaein 2017 42 Network slicing is an on-going research with several challenges 42 Isolation, Sharing, Customization Satisfy requirements of slice owner and operator/infra. provider
Conclusion (c) Navid Nikaein 2017 43 Data-driven network control is difficult 43 Reason-Predict-Control is a generic framework Prediction performance is limited by the available computing resources
(1) Can we predict user QoS/QoE per 44 application in realtime? (2) Can we learn network-user-application dependencies across various network domains? (3) Can we automatically learn the right control to apply? Open Questions (c) Navid Nikaein 2017 44
Thoughts (c) Navid Nikaein 2017 45 45
Personal Info: Email: navid.nikaein@eurecom.fr Website: http://www.eurecom.fr/~nikaeinn/ Linkedin: https://www.linkedin.com/in/navidnikaein Tel: +33.(0)4.93.00.82.11 46 Mosaic-5G.io : Mail : contact@mosaic-5g.io Website : http://mosaic-5g.io Linkedin: https://www.linkedin.com/in/mosaic-5g Twitter: @mosaic5g Contact Information (c) Navid Nikaein 2017 46