A Programmable Distributed Optimization Framework for 5G Dilip Krishnaswamy (Reliance Jio) Swaminathan S (Wipro) ICSDN 2019 Aug 10 th , 2019
A Programmable Distributed
Optimization Framework for 5G
Dilip Krishnaswamy (Reliance Jio)
Swaminathan S (Wipro)
ICSDN 2019
Aug 10th, 2019
Food for thought?
“The Americans have need of the telephone, but we do not. We have plenty of messenger
boys.”
- Sir William Preece, Chief Engineer, British Post Office, 1876
“When wireless is perfectly applied the whole earth will be converted into a huge brain,
which in fact it is, all things being particles of a real and rhythmic whole. We shall be able
to communicate with one another instantly, irrespective of distance. Not only this, but
through television and telephony we shall see and hear one another as perfectly as though
we were face to face, despite intervening distances of thousands of miles; and the
instruments through which we shall be able to do his will be amazingly simple compared
with our present telephone. A man will be able to carry one in his vest pocket.”
- Nikola Tesla, 1926
What is 5G?
Speed of >1
Gbps, latency
1 ms
Closed
control
loops,
Cognitive
Able to support
eMBB, mMTC,
URLLC use
cases
Programmable
(SDN/NFV)
NR,
disaggregated
RAN
Flexible and
context-aware
Outline
Pre-
requisit
es
Orchestrati
on
Software-
defined
RAN
Self
Organizing
Networks
(SON)
Network
Slicing
Optimization
SONNetwork
Slicing
Orchestration – Multi-layered
Reference: https://sdn.ieee.org/newsletter/september-2016/orchestration-and-control-solutions-in-5g
Simplified orchestration architecture
End-to-end/Service level
Domain/Network segment level
Network Functions
5G RAN Functional Decomposition (Disaggregation of RAN)
Reference: “NGMN Overview on 5G RAN Functional Decomposition”
5G RAN – Simplified Network Architecture
Reference: https://wiki.onap.org/download/attachments/10784151/ONAPNetworkSlicingv3_25Jan2018.pptx?api=v2
SD-RAN: RAN Architecture based on SDN principles
Reference: Paul Arnold, Nico Bayer, Jakob Belschner, Gerd Zimmermann, ”5G Radio Access Network Architecture Based on Flexible Functional Control / User Plane Splits”, European
Conference on Networks and Communications (EuCNC), 2017.
SDN-based RAN architecture Fully centralized CP
Self-Organizing Networks (SON)
Examples:
• PCI assignment
• ANR update
Examples:
• Mobility Load Balancing
• Mobility Robustness Optimization
Examples:
• Re-configurations
• Resource isolation, re-
initialization
Reference: https://www.reuters.com/brandfeatures/venture-capital/article?id=102160
C-SON, D-SON and H-SON
Centralized SON Distributed SON Hybrid SON
Reference : http://tec.gov.in/pdf/Studypaper/Self%20Organising%20Network.pdf
PCI and (C)ANR optimization
• Physical cell ID (PCI) is a locally unique integer ID in 4G/5G
o Used for handoff and physical layer design
o PCI ranges are LTE: 0-503, 5G: 0-1007)
• Objective is to assign PCI values to avoid PCI collision and PCI confusion
• Automated Neighbor Relations (ANR) builds & maintains neighbor relation tables in eNodeB and gNodeB
o Distributed ANR in eNB (4G), gNB (5G)
o Centralized ANR optimizes neighbor tables based on global information
• PCI needs to be re-optimized if neighbor relationship changes, (e.g. during distributed ANR function in RAN)
• PCI (re)allocation is an optimization problem with objective to allocate PCIs s.t. collisions and confusions are minimized with additional constraints for e.g.:
o Use minimum PCI values
o Change minimum number of assigned PCI values
• PCI optimization could be done jointly with ANR actions (blacklisting a neighbor from HO)
Reference: https://wiki.onap.org/download/attachments/68539437/OOF_PCI_Dublin_POC_Demo_20190724_v_2.0.pptx?version=1&modificationDate=1564034905299&api=v2
(Closed) Control Loop concept
• Key element of orchestration for automation of network operations
• Service/Slice Lifecycle Management (LCM)
o Design
o Instantiation
o Scaling (up/down)
o Modification
o Healing
o Termination
• Resource Orchestration
• Allocation
• Scaling (up/down)
• Re-allocation
Reference: https://wiki.onap.org/download/attachments/68539437/OOF_PCI_Dublin_POC_Demo_20190724_v_2.0.pptx?version=1&modificationDate=1564034905299&api=v2
Introduction to Open Network Automation Platform (ONAP)
ONAP Dublin Architecture
OSS / BSS / Other
Run-Time
External APIs CLI
SharedServices
Audit (POMBA)
Optimization (OOF)
Multi-Site State (MUSIC)
Logging
Microservice Bus (MSB) / Message & Data Routers (DMaaP)
Design-Time
Catalog
O&M Dashboard (VID) Interfaces
Service Design & Creation (SDC)
Du
blin
Legend
AuthN/AuthZ (AAF)
External Systems
Network Function Layer
Hypervisor / OS Layer OpenStack Commercial VIM Public Cloud
…
Third Party Controllers
Kubernetes
VNFsPNFs
PrivateEdge Cloud
PrivateDC Cloud
IPMPLS Man
aged
En
viro
nm
ent
sVNFM EMS
ONAP Shared Utilities
CCSDK
Model Utilities PublicCloud
& Others … …
TOSCA Parser
Portal Use-Case UI
ONAP Operations Manager (OOM)Manage ONAP
DCAE Design Studio
xNF Onboarding
Service/xNF Design
Workflow DesignerController Design
Studio
Infrastructure
Adaptation
(Multi-VIM/
Cloud)
Virtual
Function
Controller
(VFC)
Data Collection
Analytics &
Events (DCAE)
Correlation
Engine (Holmes) SDN
Controller
(SDNC)
Application
Controller
(APPC)
Active & Available
Inventory (AAI)
Service
Orchestration (SO)
Policy
FrameworkExternal System
Register (ESR)
Control Loop
Automation (CLAMP)
VNF SDKVVPVNF Validation
Reference: https://wiki.onap.org/display/DW/Approved+Dublin+Architecture
ONAP-based control loop for SON
• SON Control Loop (CL)
• ONAP: Open-source platform,
with basic open-source code
• Companies can use framework
to add proprietary SON
solutions
• OOF-PCI use case Casablanca o First ONAP SON PCI use case
o PoC Demo in Dec 2018
• OOF-SON use case Dublino Added SON function: ANR
o More SON data flows: FM, PM
o PoC Demo in Jul 2019
Reference: https://wiki.onap.org/download/attachments/68539437/OOF_PCI_Dublin_POC_Demo_20190724_v_2.0.pptx?version=1&modificationDate=1564034905299&api=v2
ONAP Dublin: PCI and C-ANR use case realization
Reference: https://wiki.onap.org/display/DW/OOF-PCI+Use+Case+-+Dublin+Release+-+ONAP+based+SON+for+PCI+and+ANR
“Distributed” PCI optimization
DU1
RU11
RU12
RU1k
Central
Orchestrator
CU1
DUn
RUn1
RUn2
RUnm
CUn
DU1
RU11
RU12
RU1k
Central
Orchestrator
CU1
DUr
RUr1
RUr2
RUrm
CUr
DU1
RU11
RU12
RU1j
CU1
DUp
RUp1
RUp2
RUpq
CUp
Cluster 1 Cluster v
Distributed instance of optimizer (e.g. at edge cloud, MEC)
Central co-ordinator, conflict resolver (e.g., at Domain Orchestrator, E2E Orchestrator)
Option 1 Option 2
“Distributed” PCI optimization – Research & Deployment challenges
• Placement of local optimizers – edge cloud, ‘cluster’ of cells/Dus
• Co-ordination across local optimizers
- Addressing conflicts
- Cells on the ‘edge’ of a cluster, i.e., impacting optimization in adjacent clusters
- Assignment of allowed PCI values, esp. for edge cells
- Parallel execution vs sequential execution, synchronization aspects
• Convergence of solution
• Stability & robustness of solution (how sensitive it is to even small changes)
• Central co-ordinator versus peer-level co-ordination
• Multi-vendor scenario – interoperability, hybrid scenarios
• Use of AI/ML techniques
Network Slicing – Concept of ‘‘sharing’
Reference: 3GPP TS 28.530
1
3a
3b
2
NSI Network Slice Instance
NSSI Network Slice Subnet Instance
(Simplified) Flowchart for network slice “allocation”
Instantiate a service
Determine the appropriate
NS Template
Determine whether to reuse
existing NSI or create one
Reuse?
Go to NSSI flow for each
NSSI
Map the
service to
existing NSI
Request for a NSSI
Determine whether to reuse
existing NSSI or create one
Reuse?
Determine resources to be
allocated
Return details of
existing NSSI
Assemble NSSIs and create
new NSI for the service
Allocate & configure NSSI
and return details
Yes
No
No
Yes
NSI allocation flowNSSI allocation flow
(to be repeated for each slice sub-net)
1
2
3x
4x
Step Optimization Problem
Map a service request to an appropriate Network
Slice Template(s)
Determine whether to ‘reuse’ an existing Network
Slice Instance or instantiate one
Determine whether to ‘reuse’ an existing Network
Slice Sub-net Instance or instantiate one
Determine the resources to be allocated for the
Network Slice Sub-net Instance
Network slice ‘allocation’ – optimization problem
1
2
3
4
Centralized
Multi-layered
optimization
problem
May require
iterative
approach
NP-hard
Distributed
Focus on single
layer
Local
optimization vs
global
optimization
Demarcation of
roles
Takeaways
• SDN (and NFV) enable a lot of programmability in the network
• RAN is getting disaggregated, and architected based on SDN (and NFV) principles
• Network Slicing adds another new dimension to this already interesting eco-system
• All of this throws up numerous research, architecture, design and deployment challenges
• Focusing on optimization, there are many areas, for example:
- Known SON use cases (centralized and distributed)
- New SON use cases, beyond just the RAN
- Network Slicing
- Optimal resource allocation (with constraints): multi-site, e2e perspective to be considered
Demo Steps – Add neighbors to Cell with PCI = 5
CollisionConfusion
Original neighbors After addition of 3 neighbors