journey to cloud native for 5g - openairinterface · for application consumers and kubernetes users...
TRANSCRIPT
CONFIDENTIAL Designator
1
How RAN and Core are evolving
Journey to Cloud
Native - 5G
Azhar Sayeed
Chief Technologist
2
These slides are created for the strict purpose of discussion only. Red Hat Inc. does not plan or
commit to offer any functionality described in this material as part of its portfolio of products.
Red Hat Inc. does upstream open source software development work and will continue to
participate in upstream activities and communities relative to technologies described in this
presentation. This does not imply Red Hat will or have products associated with those software
communities Nor is Red Hat committing to having such offerings in future.
Red Hat Inc. does not commit to or subscribe to any opinions, statements regarding third parties
described in this presentation. Any disclosures associated with third party companies are a
matter of conjecture and should not be taken as known facts or stated strategy of those third
parties.
DISCLAIMER
AGENDA• Background• RAN, Edge and Core Solutions for 5G
– Architecture– Requirements of Infrastructure
• Cloud Native Factors• Examples
– Customer Example– VCO 3.0
• What’s Next• Summary
3
AGGREGATION & DISAGGREGATIONThe swing of the pendulum
4
Aggregation• Main Frame• Shrink wrapped• HW/Software bundled• Closed EcoSystem• Monolithic Software
Dis-Aggregation• Intel - x86• Component integration• Un bundled• Open EcoSystem• Microservices Software
Argument for Aggregation• Speed• Performance• One Neck to Choke• Better User Experience• Support Models
Arguments for Disaggregation• Independence• Flexibility• Control of your destiny• Innovation• New Services
IN SEARCH OF A NEW EQUILIBRIUM STATE
5
Time
State
t1 t2
Equilibrium State
Equilibrium State
MonolithVertically Integrated
Multiplecomponents
Shrink Wrapped
Dis-aggregation Open-Source
NFV -VM
App - RefactoringRe-Aggregation
CNF/SDx
Value Migration
6
Hardware and Software all
bundled in one
Cables
Software
Cables
Hardware
Application
Cables
Boards
NICs
Processor
Storage
Proprietary OS
Application
Cables
Faster Boards
Faster NICs
Faster Processor ..xx
Faster Storage
Proprietary OS
Cables
Multi Function Boards
Even Faster NICs
Multi-core Processor ..xx
Even Faster Storage
Linux OS
Application
Cables
Multi Function Boards
Smart NICs
Faster and Multi-core Prcssr ..xx
Flash Storage
Linux OS
Hypervisor
Application
Cables
Multi Function Boards
Smart NICs
Multi-core Processor ..xx
Flash Storage
Linux OS
User space
Application
Application
Application
Application
Application
Application
Application
Application
Business Process/Integr
ation
7
RAN, Edge and Core
LTE RAN Evolution and 5G RAN
Mobile Broadband
Ethernet or Fiber Fronthaul
Compute Nodes at - RU (Radio Unit, DU (Distributed Unit), CU (Centralized Unit) and Edge Compute for Application in addition to packet core and DC
5G RAN
BBU/vBBU
CU
Functional Splits with LTE Radio
DU
DU
RU
RUFronthaul Midhaul
BackhaulDU
LTE vRAN
Regional or Core DCFor LTE EPC or NG-
Core (5G)Edge
Compute
Edge Compute
CU
5G New Radio
RU
PACKET CORE EVOLUTION 4G/LTE TO 5GMOVE TO CONTAINERS
9
eNB
MME
SGW PGW
HLRHSS
OCFPCRF
CP CP
DP DP
Box / Device centricLTE/4G
5G RAN4G
RAN Localized GW or Central GW Data Plane
Control Plane – Mobility, Sessions & Service Management
PGW HSSPCRF
5G - Cloud Based
OpenStack or KVM OpenStack
vBBU vMME vSGW vPGW
vPCRFvHSS/HLR
CP-DP SeparationUPF is controlled by AMF and SMFData plane extensibility
Infrastructure Requirements
11
5G Attribute Result Infrastructure Requirement - Software
FWA (Fixed Wireless Access) High Speed Broadband Service based on mmWave band spectrum
Performance, throughput – OSP and OCP, Smart NIC, FPGA support
NG-Core Next Generation Packet Core for 5G –based on Microservices and Containers
Orchestration of Containers - Kubernetes
High Frequency Spectrum with more spectrum bandwidth
High Speed Residential Broadband service => Instant Subscriber acquisition
Hardware Management, distributed computing
Massive Scale IoT Services => Hybrid Cloud Hybrid Cloud => Middleware, Messaging, API Management
Control-User Plane Separation Flexible placement of traffic exit points allows Edge compute for application optimization
Edge Compute foot prints, infra placement and management and assurance framework
Low Latency Services (URLLC) Drones, Autonomous Vehicles and Holographic calling
Low latency processing of information => RT Kernel, RT-KVM
Slicing Virtual Operators, Private Mobile Broadband, Distinct grades of service
Distributed Cloud deployment models, Infrastructure partitioning, multi-tenancy, Networking and VPN support
4x4 MIMO, Beam Forming Higher throughput per user on Macro cell and small cell
Higher Aggregate throughput => Smart NIC, FPGA etc
Flexibility, reliability, serviceability Disaggregation of software components, cloudiness
Standardized catalog, registry of components
• CNF - There aren’t many that exist today– Handful of vendors working on developing network functions to
microservices/containers• Unique requirements
– Multi-interface– Traceability– Service chaining– IP address management (flexibility)– Fast data path - acceleration– High Availability Reference Architecture– Common SDN environment– Real-Time Kernel for specialized Edge use cases– Multi-Cluster, Multi-Site, Private and Hybrid - Same experience expected
Cloud Native Network Function Challenges
12
CONFIDENTIAL Designator
Cloud Native
Design Factors
13
Factors influencing cloud native design
• Distribution of functionality, applications• Scale up and scale out => Cloudiness factors• Disaggregation and state management• Operations knowledge codification - for automation and
life cycle management• Instrumentation, telemetry and assurance• Orchestration and Management - Lifecycle and Day 2• Security
14
DISTRIBUTION & MULTI-CLUSTER ENVIRONMENT
15
Independent or stretched K8S clusters (1 node to n nodes remote sites)
Agg or Core Network
Far Edge
Edge AppvCPEStack or Multi-Tenant
RH
DU
CU
Metro Ethernet
DU
DU DU
DU Pool
Independent or distributed Kubernetes Clusters
Regional or Centralized Data Center
with NG-Core or NSA LTE Core
Near Edge
DISTRIBUTED & MULTI-CLUSTER - PRIVATE & HYBRID
https://github.com/kubernetes/community/tree/master/sig-multicluster
OpenShift Clusters c1 through c7
c1c2
c7...
Cluster Registry CRD
Single Source of Truth
Federated API
Base Federated Resources
Substitution Preferences
Substitution Outputs
Placement Preferences
Placement Decisions
Schedule and Reconcile
Auxiliary Resources
FederatedDeploymentFederatedSecretFederatedReplicaSetFederatedConfigMap
Bonus: Federate any CRD without writing code
$ oc get clusters$ openshift-install launch overrides:clusters:- clusterName: c1
replicas: 5- clusterName: c3
replicas: 10- clusterName: c7
replicas: 15
Scale Up● Exposing Hardware functionality for workload optimization (Node feature discovery, CPU
Manager)● Assigning pools of cores to specific threads vs thread pinning to specific cores
○ RT threads - deterministic behavior○ non-RT threads
Scale Out● Re-usability and load distribution
○ spawning new instances○ automatic management of component bundles - think service mesh
● Independence and abstraction from hardware/infrastructure adheres better to cloud principles
● State less
Scale Up and Scale out - CloudinessResource pinning vs dynamic allocation
17
Human operations knowledge in software - to reliably manage an application
For Builders and the community
● Easily create application on Kubernetes via a common method
● Provide standardized set of tools to build consistent apps
For application consumers and Kubernetes users
● Keep used apps up to date for security reasons and app lifecycle management
● Consume of cloud-native / kube-native applications more secure and easier
OPERATOR FRAMEWORK
Operations knowledge codification for automation and lifecycle management
EXTENDING ASSURANCE MODEL
19
Edge Collector
Edge
EdgeC
CCompute
AMQP Message Bus
Decision Engine
Regional or Central Data CenterEdge Sites
Collectdand/or
PrometheusAnsible Tower
Ansible play book Execution
PrometheusTime Series DB
AMQP Message Bus
• Automating placement of workloads that meet the criteria• Monitoring Change• Proactive and reactive action
• Rules based closed feedback loop• Requires all layers of stack to communicate• Correlation
CONFIDENTIAL Designator
Deployment/POC
Examples
20
GC Edge Clou
d
DEPLOYMENT– REAL CUSTOMER EXAMPLE
Peering
DC
The Internet
Pre-Agg Agg. IP Core
RIU
Internet
Public Cloud/OTT
NFVI SW NFVI SW
vDUvCU
SAEGW-U
vCU -Small Cell
SAEGW-C/U vIMS
PCRF, HSS, OCS
vCDNOTT CDN
CO-LO
RIU
NFVI HW NFVI HW
Apps
Partner Apps
Gi FW/NAT
MANO (VNF-M and Orchestrator)
eCPRICPRI
VCO 2.0 – ONS POC
SDRSGW MME HSS
PGW PGW
SD
VPN Internet
GiLAN
Internet
RU DUCU
Baremetal NodesManaged by OpenStack
Amsterdam Stage
Jumphost
OpenStack
CONFIDENTIAL Designator
Looking ForwardWhat’s next?
23
Unified foundation, hybrid interoperability
KUBERNETES NATIVE INFRASTRUCTURE
PHYSICAL
APP APP APP APP
VIRTUALPUBLICCLOUD
VM CC C
24
KUBERNETES NATIVE INFRASTRUCTURE ARCHITECTURE VISION
Kubernetes Container Orchestration (Apps & Infrastructure Services Lifecycle Management)
Linux Container Host on Bare Metal Servers (Immutable, Lightweight)
App
Automation
Container Registry
App & Infrastructure Services
On-premises Data Center(s)
Sof
twar
e D
efin
ed S
ecur
ity
DNS, Load Balancing
Container
VMs
App
Container
Functions
App
Service Catalog
Identity & Access
Management
Software Defined Compute Software Defined Networking Software Defined Storage
Monitoring
Cost Management
Public Cloud(s)
Container
25
Server Less Computing
Assumptions that computing, storage and networking are a commodityInfrastructure divorced from applications and servicesFunction aaS – FaaSTriggered computing
SUMMARY
• Cloud Native holds key to scale, distribution, efficiency as we build software for 5G• Careful disaggregation of components to utilize the hardware and infrastructure
without losing the cloudiness• As you make the applications and network functions cloud native, think about state
management, infrastructure independence and abstractions• Instrumentation and telemetry key to better assurance overall• Infrastructure components are more or less ready to start on boarding the cloud native
network functions (CNFs)• Careful planning wrt components drives better efficiencies - state management,
immutability etc• Contribute upstream
27
CONFIDENTIAL Designator
linkedin.com/company/red-hat
youtube.com/user/RedHatVideos
facebook.com/redhatinc
twitter.com/RedHat
28
Thank you