Towards a Carrier Grade ONAP PlatformFCAPS Architectural Evolution

Key Contributors: Ramki Krishnan, Sumit Verdi, Xinhui Li, Danny Lin, Bin Hu, Gil Hellmann, Bin Yang, Shankar Narayanan, Sastry Isukapalli, Rajesh Gadiyar, Vimal Begwani

Agenda

➢ Operational Intelligence for Dynamic Orchestration

➢ MC FCAPS Architecture – Common Data Model & Data Distribution

➢ MC Impact - VoLTE, vCPE VNF Placement (Homing) Use Case

➢ MC<->OF Interaction for VNF Placement (Homing)

Analytics Functions

Collection

Operational Intelligence for Dynamic Orchestration

Application and Infrastructure correlated context is key…

Adaptive on-demand service blue-printing and workflows

Dynamic policies using utilization intelligence, Policies using batched data

Cloud-aware workload placement using capability and resource utilization

VES

NFVI

DCAE

Dat

a D

istr

ibu

tio

n F

un

ctio

n

Multi-Cloud FCAPS

Aggregation

VNF

Normalization

Correlation

Outliers

Aggregation

Learning

Policy

SDC

Opt. Framework

…Common Data Model

Planning & Forecast

Service

Access

Virtual

Physical

Transport

MC FCAPS Architecture

• Standardized Common Data Model across Cloud Providers -Various cloud providers have disparate data structures, representations, middle-wares and more for infrastructure telemetry collection and management

• Composite NFVI Intelligence at atomic, aggregate and infrastructure capability granularities to enable cloud-aware decisioning by OF, DCAE, SO

• Enable Continuous Service Deployment with run-time resource reservations and utilization telemetry

• FCAPS Data Distribution to enable simplified and scalable many-many communications using DMaaP pub-sub

DMaaP

Replication

Active Sync

DCAE Opt. Frwk. A&AI

Ho

lmes

CD

AP

VES Collector

MSB Endpoints

MultiCloud

Infra Class - MC instance, Clusters within MC

instance etc.

Real-time Faults Metrics per Host, VM,

Network etc.

Capability, Capacity, Topology per cluster

SDN-C SDC Policy

.….OpenStack AzureVMware Wind River Kubernetes

FCAPS Common Data Model, Distribution & Integration

*Joint collaboration between VMware, Intel, AT&T, China Mobile, WindRiver

Host

Host OS

Hypervisor

Virtual Machine

Guest OS

Switching

Data Store

Resource Pool/Provider

Data Center

Capacity

Alert

Recommendation

VES Collector

NFVI Collector

DMaaP

DCAE

OF

Policy

SDN-DC

OpenStack

Azure

VMware

Wind River

Google Cloud

Mirantis

Met

rics

& In

fras

tru

ctu

re C

lass

*

EPA

Resource Reservations

Resource Utilization

Telemetry

Infrastructure Class Example

• Hardware encryption

• Hardware transcoding

• NUMA nodes available

• Storage class

• CPU, Memory, NIC class

• WAN interconnects and protocols

Cloud Agnostic Data Model

Capability

Resource Utilization

Resource Reservation • Number of tenants

• MC instances per data center

• Maximum number of VM’s

• Total number of clusters

• Total number of hosts

• Total number of CPU, MEM

• Total number of data stores

• Number of hosts currently active

• Number of running VMs

• Total number of vCPU’s powered

• VM density per host

• Available bandwidth per host

• NUMA utilization

Cloud Provider | Multi-Tenancy | Multi-Device Access | U/P Disaggregation | Business Model | Cost

Cluster Topology

• Network Fabric type- CLOS etc., 2 level, 3 level

• Interconnect BW- Max, Min

VoLTE: Distributed DC VNF Placement (Homing) Use CaseWorkflow: Continuous Deployment - Day 1 & Beyond

Current ONAP Challenge: Static MC instance selection for workload placement leading to higher cost due to

under-utilization or poor application QoE due to over-subscription of infrastructure

Value Proposition: OF to deliver the best VNF placement solution in terms of Cost, Security and Application QoE

by dynamically determining the appropriate multi-cloud instances leveraging aggregate infra data from MC

Multi-Cloud Instance 1

Enterprise

Multi-Cloud Instance 1

Multi-Cloud Instance 2

Multi-Cloud Instance 1

Multi-Cloud Instance 3

[u] – User Plane[c] – Control Plane[i] – Data Ingestion[p] – Data processing

Cloud Interconnect IPSEC VPN etc.

Edge Site 1

Edge Site 2

Edge Site 3

Core Site 1

Public Cloud DC, e.g. Azure

EPC [u] IMS [u]

EPC [u] IMS [u]

EPC [c] IMS [c]

EPC [c, u] IMS [c, u]

FCAPS, VNF [i]

FCAPS, VNF [i]

FCAPS , VNF [i, p]

FCAPS, VNF [p]

Historical Analytics

Web Conf.

Voice Translate

ONAP OF

Residential vCPE: Distributed DC Placement (Homing) Use CaseWorkflow: Continuous Deployment - Day 1 & Beyond

Physical resources at Customer Premise (BRG)

Resources in Cloud (vG)

Infrastructure resource

Per-customer resource

vG MUX

BRG

vGBRG

vBNG

vG

BRG – Broadband Residential Gateway

BNG – Broadband Network Gateway

vG – Virtual Gateway

vG Mux – Virtual Gateway Multiplexer

Logical subscriber link

ONAP OF targeted for R2 to address R1 use cases (VoLTE, CPE) and upcoming use cases (5G etc.)

ONAP OF

TunnelXConn AllottedResource:

Requirement:TunnelXConnCapability

vG VNF Resource:vG (VFC)

Policy: Latency, Capacity check• {Customer_Location, TunnelXConn} < X ms• Allotted resource has capacity for new order

Policy: Co-location{TunnelXConn, vG} in same Cloud

Policy: Capacity check• Cloud has capacity for new vG

Home

MC <-> OF InteractionKey Contributors: Shankar Narayanan, Sastry Isukapalli, Ramki Krishnan

DMaaP Replication

Policy Driven Homing

Optimized Homing recommendations

A&AI Available

Cloud/service instances

Decomposed Service components (Demands)

Cluster Capabilities, Capacity, Topology

Infra Class - Resource Utilization/Reservation

Multi Cloud (MC)

Homing ConstraintsDistance

Regulatory

Latency

Security

Proximity/Colocation(virtual & physical)

Diversity (Disaster Zones)

Bandwidth

Cost (Provider)

Latency (Customer)

Distance (Customer)

Load balancing (Provider)

Cluster HW Capabilities(SRIOV, Hardware

encryption, Transcoding, NUMA boundaries)

Site Reliability

Cluster SW Capabilities (Availability zones,

Affinity/Anti-Affinity)Aggregate Utilization (Tenant, Hosts etc.)

Aggregate Capacity (Tenant, Hosts etc.)

Capacity check

Quota

Service Requirements

Runtime Metrics

Optimization Objectives

Runtime Queries

Cloud Capabilities

HOMING (OF-HAS)

Constraint Model plugins

Cloud Agnostic Std. Data Model from MC, Other plugins

from SO to SO

from Policy

Constraint Categories

Algorithm plugins

DCAE

OF Model-Driven Approach

Data Sources

SDN-C

PolicyEngine

AAI

DCAE

Translators

Interfaces

Model Specification

DataSpecification

SolutionEnvironmentApplication-

Specific API

OptimizationResults

Goal: provide a platform that facilitates model-driven optimization

• Model-driven (Declarative)• Library of building blocks

(as SDC models, policy models, etc.)• Reuse models from OF-contrib library

• Recipes for model composition• Adapt at operation time (no new code)

(new constraints, objectives, runtime flags)

• Op-ex benefits – reducing software dev costs• Rapid analyses – what-if scenarios via config

• Platform can seamlessly provide new functionality/advances to application

Multi Cloud

Constraint Model

Repository

ONAP-OF Based on Standardized Constraint Modeling

Global Constraint Catalog

Chuffed

Data(dzn format)

Flat Zinc

MiniZinc Model

MinizincStandard LibraryAvailable

Technology

AvailableExtensions

StochasticMinizinc

Contributed Models

MiningZinc MiniBrass

Constraint-Based Mining

Soft ConstraintsUncertainty Considerations

ONAP-OFContributions

OperationalEnvironment

DataInterfaces

Model/ConstraintTranslators

BuildingBlocks

Recipes andKnowledge Base

LibMzn(Embeddable

Library)

BACKUP

Example Problem: Budget-Constrained Max Flow

Example Problem: Budget-Constrained Max Flow

Data Template

Model

Constraints from model designer

Constraints from developeror from service provider/vendor

(can be from SDC or Policy Engine)

Macros from OSDF

Templating Support from OSDF

Data from A&AI, SDN, DCAE or Multi-Cloud