2. Our speakers
Srilakshmi Valisammagari
Senior Technologist and Strategist
Verizon Enterprise Solutions
Jim Fagan
Director, Global Platforms
Telstra
David Hughes
VP, IP Engineering
PCCW Global
3. Outline
• Closed-loop automation is the end goal for NFV.
• Telcos are on a journey from manual processes to writing scripts
(Python), building models (YANG, TOSCA), introducing model-driven
programmability, analytics and finally AI.
• With machine learning the network can detect or predict faults, know
how to fix it (e.g. re-route traffic, spin up a new instance) and execute
it without any human intervention.
• However, in the early days it may be risky and many CSPs will want
tools that will let them confirm changes are handled correctly.
• This panel will discuss the state of play for closed loop automation in
NFV and what needs to be done to make it a reality.
4. Control Loop Automation
• Open loop systems capture telemetry and diagnostics information
from the underlying infrastructure (syslog, SNMP, fault and
performance events), conduct analysis, and provide reports or alarms
to the operations team.
• Closed-loop systems continuously monitor the system for problems
(fault, performance, security, etc.) and compute a set of signatures
based on detected anomalies.
• A policy engine then interprets these signatures and recommends
corrective actions to repair the system.
• Once the system has been repaired the status is monitored to check
the problem has been fixed.
• The goal is to minimise downtime
6. ETSI OSM view of NFV automation
• Design and development (onboarding) – full automation still some
way off
• Instance lifecycle – what can be changed in-life and what is
immutable?
• In-life
• Performance – auto scaling
• Resilience – fault tolerant load sharing
• Service monitoring, assurance and reporting
• All this automation is policy driven
7. ML for Closed-Loop Automation
• Rather than performing reactive analytics, ML can help predict
anomalies in advance based on time-varying signals.
• Instead of applying hard-wired policies or rules to determine next
best action, ML can make the recommendation more intelligent
based on data, context, patterns and outcomes.
• As more data is collected over time, ML can learn from successes and
failures and adjust models to reduce errors. ML can make the system
more dynamic and proactive.
• Opportunities for ML: Self-Optimizing Networks; cybersecurity &
threat analytics; fault management; customer experience
management; traffic optimization.