NFV Made Efficient: Living at the Edge

Presented by :

Today’s Presenters

Moderator

Simon StanleyAnalyst at LargeHeavy Reading

Bob Monkman Sr. Enterprise Marketing Manager, Networking Software Strategy & Ecosystem Programs, ARM

Karl Mörner Senior Vice President Product Management, Enea

Agenda

• Introduction

• Choosing the Right Architecture for Virtualized Edge

• Containerized VNFs

• Solutions for vCPE and the NFV Edge

• Q&A

Source: Cisco VNI Global IP Traffic Forecast, 2016–2021

• Demand growth & shifts:– Rapidly growing user traffic– Migration to cloud services– Global IP traffic doubling by

2020– Operator demands for network

flexibility, agility and optimizedcost

• Key trends:– Network virtualization– Containerization– Wider use of open source solutions– New ecosystems

Market Drivers

Network Virtualization

Server Hardware

White Box Switches

VirtualApplications

Network Services

• Enables rapid service provisioning and lower capex/opex

• Software Defined Networking (SDN)– Abstracts and automates provisioning

– Separates data and control planes

• Network Functions Virtualisation (NFV)

– Virtual Network Functions (VNFs) on common hardware

• Expanding VNF Ecosystem

• Open initiatives

– OPNFV, ONAP etc.

NFV Infrastructure (NFVI)

Source: Earlswood Marketing

Enterprise vCPE

SOHO vCPE

• Central data centersand edge computing

• Different performance requirements– Cloud Servers– Local Servers– Enterprise vCPE– SOHO vCPE

Internet

Cloud Application

Servers

Local Application

Servers

NFVI

2020: More nodes, more needs

1msend to end

3.7exabytes

per month

22xbandwidth

increase

~30xaccess nodes

Operators: “I need more capacity and lower latency within the constraints I have.”

© ARM 2017

The power of a differentiated ARM ecosystem

Shippingtoday

4-64core

solutions

Up to

7xefficiency

demonstrated

© ARM 2017

Cortex-A72: Ideal for dense compute environments

Cortex-A72 is <20 %

size

Single Broadwell CPU + 256K1 L2

~8mm2

Cortex-A72 MP4 + 2MB L23

~8mm2Single Cortex-A72

core 2 ~1.15mm2

A quad core Cortex-A72 with 8x L2 cache RAM is

the same size

1Source: Estimated from die-shot image provided by Intel at IDF 2014. 2/3Source: ARM trial implementations on TSMC 16FF+, using ARM Artisan libraries

Core

© ARM 2017

Workload Efficiencies- Accelerators

ARM Ecocsystem SoCs deliver a range of accelerators/offload

Packet Processing

Crypto functions

Traffic management

Full IPSec Offload

Virtual Switching Offload

Network Virtualization protocols

Delivering proven workload efficiencies v. running on the CPU

Up to 10x performance

Up to 80% reduction in jitter

Dramatically lower latencies

© ARM 2017

Workload Efficiencies: VNF Fastpath Advantage

0

1

2

3

4

5

6

7

Linux Fastpath

Fastpath Speedup (Observed)

IPv4 forward NAT Ipsec

Fastpath Speedups > 6x

NXP processors provide

complete HW assist of

virtualized I/O (incl L3 &

security)

6x

Rel

ativ

e P

erfo

rman

ce (

No

rmal

ized

to

Ho

st)

Network Fastpath Offloads are Sufficient to Overcome Significant Virtualization Overheads

CPU core load < 1%

ProvidedCourtesy of

© ARM 2017

Space Efficiency- The World’s Tiniest Pharos Pod

Fully Functional Pharos Pod @ ~1 cu. ft.

1 Jump Server, 3 Controllers & 2 Compute

Nodes

12 port Switch, Power Supply, Fans

Marvell MACCHIATObin boards w/ Marvell

Armada 8040 SoC (Quad ARM Cortex-A72-

based)

Dual 10G NICs, 16Gb RAM, PCIe, SATA & more

Runs Danube NFVi – sample VNFs

“Data center on a Desktop” dramatically improves the accessibility of OPNFV NFVI for Developers

Check it out in ARM Booth G1

© ARM 2017

Introducing the world’s tiniest OPNFV Pharos Pod

Typical Pharos Pod occupies 20U Rack

min. >20 Cu. Ft.

NFV PicoPod is approx. 1 Cu. Ft.

Pre-Order @ https://www.picocluster.com/products/nfv-picopod

© ARM 2017

• Early SDN technologies based on HW virtualization in OPNFV

• OpenStack management of KVM-based VMs for VNFs

• Higher overhead (memory, performance, complexity) approach

• Consensus building that the Virtualized Edge favors a lighter weight solution

• Many are now looking to leverage lighter weight OS virtualization

• Docker, CoreOS

• Kubernetes, Mesosphere, Weaveworks etc…Microservices

• OpenStack, SDN Controllers may still play a higher level role in these configurations

• ARM Ecosystem is now following the OpenRetriever project forContainer-based support

• And building PoCs to demonstrate the strengths of microservices approach

The Right Architecture for Virtualized Edge

© ARM 2017

Container PoC project: Why Containerized VNF?

Compared to NFV+VM

approach, we anticipated:

Lightweight footprint

Efficient resource consumption

High density deployment

Fast deployment

ContainerContainerContainer

Orchestration & Automation

© ARM 2017

ARM based OPNFV Container PoC

Jump Server

VPN

Firewall

VPN

Gateway Router

POD

3x Compute Nodes

2x Controller Nodes

© ARM 2017

PoC – Containerized VNFs on ARM NFVi- Networking Services Deployment

Virtual

Machine

Virtual

Machine

Master

Virtual

Machine

Bare-Metal

Machine

OpenStack on AArch64

Minion#1 Minion#2 Minion#3

VNF VNF VNFVNF

ARM-based OpenStack as VIM

OpenStack Magnum as container

service

Kubernetes as COE

Flannel plugin for Kubernetes as CNI

Containerized VNFs within a VM and

bare metal, both

© ARM 2017

Demo Scenario – Deliver Network Services with vCPE for Enterprise Branch

vCPE: Containerized OpenWRT

MAN

Enterprise BranchIPsec Tunnel

Service Provider’s Data Center

using standard ARM Servers

Enterprise Branch with No

Internet Access

Thin CPEVNF VNF VNF VNF

vCPE

Enterprise Branch Connects

to Internet

• When are you likely to will use Container technology in your developments?

– Already using

– During next 12 months

– 1-2 years

– 2-4 years

– Currently evaluating

– No plans

Audience Poll #1

Enea developing solutions for a connected society

▶ Enea already holds a world-leading position on this

market with proven high-performance solutions.

▶ Networks based on Enea’s Operating Systems currently

serve more than 3 billion wireless subscribers globally.

▶ The world’s leading telecom companies turn to Enea

when building global communications infrastructure.

Enea in brief

Head Quarter

Kista, Sweden

Revenue

501.3 MSEKOperating Margin

23.7 Percent

No. of Employees

464

R&D OPEX

Investment

18 Percent

Listed on

NASDAQ Stockholm

Numbers for 2016

The vCPE at the NFV Network Edge

Orchestration

vCPE

Enea did the initial bring up of the OPNFV

architecture for ARM hardware and delivered

the first “Arno” release in H1 2015

Enea demonstrated together with ARM a first

“multi-vendor” demo of the OPNFV

architecture at the ARM Partner Meeting in Q3

2015.

Enea was one of the founding participants of the

ARM Band project in OPNFV in Q3 2015

Enea ported OPNFV to the NFV PicoPod in

collaboration with ARM and Marvell Q2 2017.

Enea launched in Q4 2015 the world’s first

ARM-based NFV lab compliant with the

“Pharos” specification from OPNFV

Enea has brought up the Brahmaputra,

Colorado and Danube releases of OPNFV on

ARM hardware

Enea today hosts a multi-architecture NFV lab

with 10+ pods

22

Key Enea’s open source NFV contributions

1st

1st

1st

The vCPE at the NFV Network Edge

Orchestration

vCPE

Drivers at the edge of the network

▶ Shorter time to revenue

▶ Service deployment flexibility

▶ No vendor lock-in

▶ Lower OpEx

▶ Lower CapEx

vCPE Deployment Scenarios

PNF

PoP/Central Office

uCPE white box

NFV compute

VNF VNF

Customer Premise

COTS Server

NFV

VNF VNF

PoP/Central Office

pCPE

Customer Premise

uCPE white box

NFV compute

VNF VNF

Customer Premise

COTS Server

NFV

VNF VNF

PoP/Central Office

Customer premise only1

PoP/Central office only2

Distributed3

Optimizing the edge of the network

EMS

VNF VNF VNF

EMSEMS

Network

virtualization

Storage

virtualization

Compute

virtualization

Virtualization Layer

NetworkStorageCompute

OSS/BSS

Orchestrator

Virtualized

Infrastructure

Manager

VNF

Manager

NFV CPE

OPNFV compute and control

Op

en

Sta

rdard

Pro

toco

ls an

d A

PIs

BOM optimization

NFV feature complete

Building on standard protocols and APIs

Standard VNF APIs

Standard integration Orchestration APIs

Flexible and standard VNF Lifecycle APIs and Protocols

VNFsVNFs

Orchestration

11

23

3

OPNFV compute and control NFV CPE

1

2

3

Building a fast virtualized networking solution on open source

Acceleration techniques Architecture

OVS optimizations

Hardware acceleration

Core partitioning optimizations

OpenFastPath

White Box Hardware NIC

NFV CPE

VNF space

NFVI

Open vSwitch-DPDK/ODP

DPDK/ODP

DPDK/ODP

QEMU/KVMContainers

OpenFastPath

Application

SR-IOV

Containers vs KVM

Characteristics drive the need for mixed containers and VMs

VNF Networking performance is not the main differentiator between containers and VMs

Container virtualization

▶ Small footprint

▶ Fast boot/start

VM virtualization

▶ Contained upgrade

▶ Secure

Mixed VM and

container VNF

hosting

OVS Internal Bridge/Vhost-user

Platform

Container VM

VNF VNF

Container

VNF

Orchestrating the Edge

NFV CPE

VNFVNF VNF

NFV Platform

Orchestration

Carrier Edge

PoP/COCustomer Premise

Adaptable Orchestration with Multiple Interfaces

• Interfaces for centralized VNF Management and Service Function Chaining

• NETCONF/REST

• Container and VM virtualization

• Small footprint and CPU utilization

• Standard protocols

• Direct Container access

• Container spefici virtualization

• Small footprint and CPU utilization

• Integrate with OpenStack or Orchestration

• OpenStack

• Container and VM virtualization

• Full OPNFV/OpenStack integration

• Larger footprint

• Integration points

• Orchestration

• VIM

1

2

1

2

VNFVNF VNF

Enea NFV Portfolio

Central Office

Data Center

Customer Premise

Enea NFV Access

VNFsVNFs

Orchestration

Enea NFV Core

Enea NFV Core

NFVi + VIM + SDN

Enea NFV Core

Virtual computing Virtual Storage Virtual Network

Data Plane

vSwitchDPDK

OpenStack

Ceph OpenDaylight

High Availability

KVM

Virtual Network Functions

NFV software platform designed

and optimized for point-of-

presence / central office

Enea NFV Core Key Values

High Availability

▶ Feature complete OPNFV

Doctor implementation

▶ Controller node redundancy

▶ Compute node redundancy

Commercial Grade

▶ Quality Assured using OPNFV

test suites, self hosted arm and

x86 Pharos Lab and in-house

QA

▶ Performance verified using

OPNFV benchmarks

High Performance

▶ First class virtualized

networking performance

▶ Performance tuned Enea

vSwitch (OVS-DPDK based)

and KVM

ARM and x86

▶ Verified and optimized for

ARMv8 Cavium Thunder X

▶ Verified and optimized for Intel

Xeon

Enea NFV Access

Enea NFV Access

Data plane

VNF management

virsh

OVS DPDK

KVM/QEMU Docker

OFP

Virtualization Containers

Virtual Network Functions

Edge linkorchestration

interfaces

OpenStack API

Rest

CLIAPT packet

management

FCAPS

Docker API

NETCONF

VNF to VNF

interconnect

OVS-DPDK

Vhost-user

NFV software platform

designed and optimized for

customer premise

Enea NFV Access Key Values

Customer Premise

▶ Minimized footprint

▶ High compute density

▶ Container and VM support

No Lock-in

▶ Multiple VNF lifecycle

management interfaces

▶ Open APIs and standards for

whitebox deployment

▶ Device management support

for FCAPS

High Performance

▶ High networking performance

with optimized vSwitch

▶ Fast boot speed

ARM and x86

▶ Scales from 2-core ARM edge

device with single NIC to high-

end x86 servers

Customer premise

Enea NFV Deployment Scenarios

Case 2: Orchestrator + OpenStack

NFV Core

Orchestrator + VIM

Edge PoP/Central office

Case 1: Orchestrator only

OpenStack/

REST/

Netconf

Network traffic tunnel

VNFOpenStack APIs

VNF

Customer premise

NFV Core

VNF

Orchestrator + VIM

NFV Access

Edge PoP/Central office

OpenStack

APIs

Network traffic tunnel

VNFVNFOpenStack APIs

VNFVNF

NFV Access

VNF

Enea NFV Core and Access support Service Function Chaining (SFC):

▶ within each platform

▶ between the platforms

• Are you building or deploying ARM based vCPE solutions in the near future?

– Already started

– Within 0-12 months

– Within 12-24 months

– Considering vCPE options

– No plans

Audience Poll #2

© ARM 2017

Summary

The somewhat competing challenges of evolution to 5G &

Virtualization will require optimizations for efficiency

The ARM ecosystem is delivering compelling innovation in

efficiencies for space, power, performance and cost

Enea delivers significant value in commercially-backed software for

NFVI platform, as well as management & orchestration, edge to core

Coupled with custom integration and support services

© ARM 2017

Announcing…ARM Infrastructure Developer Ecosystem (AIDC)

https://developer.arm.com/aidc

© ARM 2017

Founding Members

Questions and Answers?

Moderator

Simon StanleyAnalyst at LargeHeavy Reading

Bob Monkman Sr. Enterprise Marketing Manager, Networking Software Strategy & Ecosystem Programs, ARM

Karl Mörner Senior Vice President Product Management, Enea

Thank you for attending!

Upcoming Light Reading Webinars

www.lightreading.com/webinars.asp

www.arm.com