Denis Murphy
Offering Management
Automating the journey to Cloud Native Intelligent Networks
“Ultimately 5G is about the move to software at the center of the network”
2-Appledore Research Whitepaper, Take a Revolutionary Approach to 5G Cloud Native, June 2020 2
The modernization of networks is key for 5G success and overall network transformation.
Managing environments with both physical & virtual resources
Lack of devops capabilities and skills needed to innovate
Lack of end-to-end orchestration capabilities
Comprehensive full stack automation
Hybrid solution spanning all resources
Adopting a cloud operating model
Challenge Opportunity
4
Poll 1
What do you see as the key drivers for network automation in a cloud native network?
1. Reducing time to market for new services
2. Improving ability to adapt to changing business conditions
3. Reducing OpEx
4. Reducing creation time for faster service innovation
Cloud networking and services for large scale NFV & 5G deployments are complex to manage
Edge Cloud
Micro Cloud
Edge Cloud
Edge CloudEdge Cloud
Micro CloudMicro Cloud
Micro Cloud
K8s restarts broken POD and notifies CP4NA
CP4NA auto configures the restarted POD and notifies any related CNFs of any changes to maintain service
CP4NA manages Network Services and VNFs across heterogeneous Clouds
Complex dependencies across multiple technology layers
OpenStack
Network Fabric
Hardware - CPU, Memory, Acceleration, FPGA
Kubernetes
Hardware design/tuning & VM/Container Placement within a site
Viewpoints
CNF Type PODsVNF Type VMs
internal virtual machine/container
placement, hardware
dependencies and configuration
Logical network design and xNF stitching across sites
VNF1 CNF1VNF2 CNF2
site placement and connecting external
logical interfaces
Network Service point of view
• What site do we put which version of a VNF/CNF?
• How do we fit everything we need for expected performance across these sites and WAN links?
• Figure out which existing network services to bind to?
• How to upgrade from one flavour of a network service to another with no down time?
• When to move an xNF from one site to another?
Hardware tuning point of view
OpenShift NFVIOpenStack NFVI
NodeNode
Network Switch Network Switch
NodeNode
Compute: CPU, Memory, Acceleration & FPGA
Network Switch
RHEL Host
Network Function VM
Network Function VM
Compute: CPU, Memory, Acceleration & FPGA
Network Switch
RHEL
Network Function POD
Network Function POD
RHV
RHEL Guest RHEL GuestContainer
Networking
Telemetry
System Config
OpenShift Config
Hardware accelerated
Provider Networking
TelemetrySystem Config
OpenstackConfig
DefaultHardware accelerated Default
xNF VMs/Containers require specific hardware and tuning to run in a performant manner (or at all).
Tuning parameters include • BIOS settings• NIC parameters• Hypervisor parameters• Operating System kernel
parameters• FPGA parameters
Container VIM Linux Kernel needs to have non-conflicting drivers and modules, e.g. • NIC drivers• Protocol Stacks, e.g.
SCTP/GTP• Container Networking
plugins
9
Poll 2
What are the main actors that inhibit, or slow down, the adoption of network automation today
1. High learning curve due to complex technology and multi-vendor environments
2. Resistance of people to process and skill change due to automation
3. Limited ROI visibility/measurability
4. Upfront expenses
Apply Cloud Native techniques and machine enabled automation
Lifecycle Automation
0% 100%
xNF Standard Lifecycle Cloud based tool chainIntent driven Orchestration
• Wrap xNFs with a self contained operational lifecycle
• Natively onboard autonomous CNF Operators
• Focus on modelling the network service rather than programming lifecycles
• Auto reconcile network services to cope with planned and unplanned xNF changes
• Tools to enable automated onboarding and testing of xNFs and network services
• Self service network service/slice design and behaviour testing
11
Normalized Lifecycle Modeling
• Maintaining different skills, methods & procedures for different vendors is expensive and slows down innovation
• Approach:
o Reduced complexity through standardized operational lifecycle and tools – standardize for consistency
o Common operational lifecycle for each xNF & Service
o Generate complete Network Service lifecycles using relationships and opinionated patterns
• Outcome:
o Increased levels of operational automation by modeling the end-to-end service with runtime operational requirements in mind
Standardized operations for all xNFs to enable consistent model-driven automation with CI/CD toolchains
12
Intent Driven Orchestration
• Manual programming is time consuming, error prone and inefficient
• Approach :
o Design with intent using declarative based models
o Model the service rather than program its lifecycle workflows
o Auto-generate & execute the most efficient steps
o Reconciles the actual and target state of all network cloud stacks
• Outcome:
o Intended Operational State of complex Service maintained automatically
Models the desired service operational state rather than pre-programming workflows
13
Service Design & TestingAutomation for the service itself and underlying resources for test, pre-production, and production environments
• DevOps approach is required to automate Network Cloud operational process, reducing complexity and manual effort
• Approach:
o Integrated design and test framework
o Quickly onboard xNF components into an automated CICD lifecycle
o Gain lifecycle visibility before deploying
• Outcome:
o Reduces service design time by up to 80% and diminishes the Network Service operations cost by up to 60% - all while reducing CAPEX upwards of 10%
14
Dynamic Service Assurance
• Reducing time to diagnose incidents and avoid outages
• Approach:
o Embrace frequent changes with real time insight
o Fix issues before they become service and customer affecting
o Real time context for rapid resolution
o Go back in time - know what happened, when
• Outcome:
o Quickly isolate problems for faster mean time to repair
Real-time view of network and cloud infrastructures using AI to drive decision making and process automation
Collaborate
Natural LanguageProcessing
Event Processing
TopologyStore
ChatOps
Notification
Tickets, Collab tools
Events
Environment Topology
Probable Cause
Anomalies
• Event Clustering• Seasonal Analysis and Suppression• Weighted probable cause
• AI driven Model selection• Variance Analysis• Dynamic Threshold
Logs
Guide & Resolve
LogProcessing
Metrics
MetricProcessing
Anomalies
• Dynamic Dependency Mapping• Real-time and historical topology analysis• Visualize probable cause
Correlated Event Groups
Run Books
+ Unstructured dataImplementing AIOps – Structured
Closed Loop Automation
16
Poll 3
What areas do you see most benefit from infusing AI into network automation?
1. Predictive maintenance
2. Self diagnostic, automatic problem detection
3. Self-healing of networks
4. Intelligent network operations
17
Closed-loop Operations
Feedback loop of communication between assurance and orchestration to enable zero touch operations
• Networks are becoming increasingly dynamic and many applications require low latency—often, no time is available for human interactions
• Approach:
o Incident detection and resolution
o AI-driven resolution of identified errors, with further auto diagnostics for unknown errors
o Sense and Respond to issues or opportunities for optimization and select opinionated patterns to execute
• Outcome:
o AIOps to reduce operational expenses greater than 5x; Improve network visibility and customer responsiveness
Network Cloud Stack requires an SRE style operating model
Vendor A PoDs Vendor B PoDs
Fully automated delivery modelHighly scalable & repeatable for each PoD type
…
V/CNF & service definitions
Cloud deployment templates
Deployment
Closed-loop automationMaintenance
Key operating model changesFully software-defined system requires new tooling, processes & skills
Architecture Standard, open virtualization & software components
Ops tooling Intent-based (vs. workflow-based)
Process SRE (Site Reliability Engineering), DevOps
Skills Software eng + admin
Fully software defined
architecture ∞
Designing networks for the cloud with AI and automation
IBM Cloud Pak For Network Automation
Network
Service
Design
Reconcile the actual and target state
of all network cloud stacks
Set of pre-defined lifecycle types with
dependencies and relationships and a
library of opinionated patterns that address
planned and service restoration use cases.
Intent
Orchestration
AI ResolutionD
om
ain
Au
tom
ati
on
Mo
de
ls
Sense and Respond issues or opportunities for
optimisation and select opinionated patterns to execute
events/metrics
20
Poll 4
What best describes where you or your client are today on the journey to cloud-based networking?
1. No plans in place or just considering
2. Starting to plan and evaluating solutions
3. In testing/evaluation or initial deployment
4. Already have some implementation in network
Lower Costs
Improves business process and service assurance while lowering operations costs
Exploreand schedule a free virtual consultation with an expert
Get started
Deploy Faster
Accelerates the delivery of networks and services through AI-powered automation
Run Anywhere
Runs on any cloud, anywhere, and manages any network vendor infrastructure
With IBM Cloud Pak for Network Automation, you can evolve to zero-touch network operations with AI-powered automation
21
https://www.ibm.com/cloud/cloud-pak-for-network-automation