democratizing the network edge - openairinterface · democratizing the network edge focus on the...

Democratizing the Network Edge – Software Eats the Access Network –

Larry Peterson

70+

70+

2

Operator Partners

Vendor Partners

Innovator Operators

Innovator Vendors

Collaborators

Guru Parulkar, Exec Dir

Andre Feutsch - CTO & ONF Chair

Jochen Appel - VP

Amin Vahdat - Fellow

Dai Kashiwa - Director

Rob Howald - VP

Shao Guanglu - SVP

George Tchaparian - CEO

Firay Yaman Er - CSO

Nick McKeown - Prof

ONF BOARD

ONF Ecosystem – 160+ Members – An Operator Led Consortium –

Public Clouds

Users

&

Devices

Edge Cloud

It’s a Multi-Cloud World

IXP/Telco Clouds

Public Clouds

Users

&

Devices

Central Offices

It’s a Multi-Cloud World

IXP/Telco Clouds

Commoditize, Virtualize and Disaggregate the Access Network

CORD: Central Office Re-architected as a Datacenter

Public Clouds

Users

&

Devices

Access-Edge Cloud

It’s a Multi-Cloud World

IXP/Telco Clouds

On-Prem

Service Chain

Public Clouds

Users

&

Devices

Access-Edge Cloud

It’s a Multi-Cloud World

IXP/Telco Clouds

On-Prem

Commodity / White-Box Hardware – Traditional Datacenter –

WAN Routers

Switching Fabric

Compute Storage

Commodity / White-Box Hardware – CORD: Central Office Re-architected as a Datacenter –

I/O I/O

PON

, RA

N

(Acc

ess)

RO

AD

M

(Co

re) White-Box OLT (for PON)

48-port, 1RU, I/O Blade

Disruptor’s Dilemma

9

Disaggregate to Catalyze Innovation

Value proposition of open networking and SDN

Integrate to Facilitate Adoption

Requirement for operational deployment

Disaggregation – Micro-Services and SDN Applied to the Access Network –

10

11

Disaggregation Techniques

Closed &

Bundled

Control Software

White-Box Switch

mS mS mS

Programmable Forwarding Plane

mS

Decouple Network Control and Data Planes

Adopt a Micro-Services Architecture

Move Functionality into the Switch Fabric

Trellis

P4Runtime Contract

ONOS

Stratum OS Stratum OS

Stratum OS Stratum OS

Northbound O&M Interface

POD

Disaggregating the Cellular Network

Trellis

P4Runtime Contract

ONOS

Stratum OS Stratum OS

Stratum OS Stratum OS Radio Unit

Distributed Unit

Radio Unit

Distributed Unit

Northbound O&M Interface

Central Unit

POD

Disaggregating the Cellular Network

User Plane (Data)

Trellis

P4Runtime Contract

ONOS

Stratum OS Stratum OS

Stratum OS Stratum OS Radio Unit

Distributed Unit

Radio Unit

Distributed Unit

Northbound O&M Interface

User Plane (Control)

CU Control

RAN Control Central Unit

Control Plane

POD

Disaggregating the Cellular Network

User Plane (Data)

User Plane (Data)

Trellis

P4Runtime Contract

ONOS

Stratum OS Stratum OS

Stratum OS Stratum OS Radio Unit

Distributed Unit

Radio Unit

Distributed Unit

Northbound O&M Interface

User Plane (Control)

CU Control

RAN Control Central Unit

Control Plane

POD

Core Control Plane

User Plane (Control)

Disaggregating the Cellular Network

Video Archive

Video Archive

Multi-Cloud Service Mesh

16

CDN CDN

SPGW-u SPGW-c

MME HSS

CU

CDN Video

Archive

To Internet

Edge Services Cloud Services

GCP

Access Edge

RU

DU

RU

DU

Integration – Configuring and Controlling an Integrated System –

17

Disruptor’s Dilemma

18

Disaggregate to Catalyze Innovation

Value proposition of open networking and SDN

Integrate to Facilitate Adoption

Requirement for operational deployment

Automate the Operationalization of Disaggregated Components

Declarative Intent / Avoid Hard-coded Dependencies

Centralized Authority / Avoid Balkanized Control

Stratum OS Stratum OS

Stratum OS Stratum OS Radio Unit

Northbound O&M Interface

Distributed Unit

Radio Unit

Distributed Unit Control

App …

Kubernetes

mService … mService Control

App

ONOS

CI/C

D To

olch

ain (XO

S)

Platform

POD

Profile

Approach to Operationalization

Approach to Operationalization

20

Declaratively specify system composition and operational behavior

Declarative Modeling Language

Generate code to operationalize: APIs, state mgmt, enforce invariants

Generative Toolchain

Define a set of abstractions to help manage complexity

Core Models

The resulting system is self-contained

All necessary authoritative state is local

Includes sufficient tools to monitor operational state

Research Opportunity

21

Edge Native Applications

Wearable Cognitive Assistance • Like Google Maps, but with additional sensors (other than GPS)

Autonomous Robots • Drone racing!

22

Challenge of Edge Computing

What unique capability does the mobile access network offer? • Low-latency proximity to end-users

• Intrinsic support for mobility

Challenge of 5G is to Simultaneously Support… • Low Latency – Moving functionality to the edge, closer to devices

• Mobility – Accessing that edge functionality while continuing to be mobile

• Customization – Both functionality and differentiated service

23

Mobile Broadband (2G – 4G)

24

Datacenter

Access-Edge

Move Functionality to the Edge

25

Datacenter

Access-Edge

Mobile Cloud (5G)

26

Datacenter

Access-Edge

Mobile Cloud (5G)

27

Datacenter

Access-Edge

EDGE CLOUD

TELCO CLOUD

PUBLIC CLOUD

OWNED BY EDGE CLOUD

PROVIDER

OWNED BY CONNECTIVITY

PROVIDER

OWNED BY PUBLIC CLOUD

PROVIDER OWNED BY

OTT SERVICE PROVIDERS

Multi-Tenant Edge

– One Company (e.g. AT&T, Amazon, Crown Castle, Google) May Play Multiple Roles

Edge Cloud Stakeholders

Factors

29

1. Cost dictates a distributed solution, with some functions running in datacenters, some in edge sites, and some on-prem.

2. Mobility inherent in 5G dictates that edge services, not just connectivity, be able to migrate to support mobile UEs.

3. Monetization dictates the need to offer differentiated services to different classes of subscribers/applications.

4. Performance dictates that functionality be implemented in most appropriate hardware (e.g., GPUs, TPUs, P4-Switches).

5. Autonomy dictates that different stakeholders will be responsible for controlling and managing different components.

Requirements

30

1. Distributed – Service meshes that span multiple clouds

2. Mobility – Service chains that migrate between edges

3. Slicing – Differentiated resource allocations and customization

4. Heterogeneous – Range of implementations choices

5. Multi-Tenancy – Multiple stakeholders (trust domains)

Call-to-Action

SIGCOMM CCR – April 2019

31

Democratizing the Network Edge

Focus on the Access-Edge: Where the Cloud and Access Collide • Requires a qualitatively new architecture

• There is an opportunity to have impact

Participate in Open Source • Commodity hardware and open source software is available

• Being embraced by both network operators and cloud providers

Operationalize Your Results • Experience with live deployments spurs research

• It’s the missing link on the tech transfer path

• CI/CD tooling is increasingly available

32

Questions?

https://guide.opencord.org

33