from sdn to cloud networking

FROM SDN TO CLOUD NETWORKING

EXECUTIVE VICE PRESIDENT, SOFTWARE SOLUTIONS DIVISION

Bob Muglia

TODAY’S REALITY Instability & Constant Change

of CEOs feel confident in

their ability to plan for growth

10%

Source: Gartner, Inc., “CEO and Senior Executive Survey 2013: The Opportunity Now Is to Drive EA From Business Priorities” March 2013

DATA CENTER EVOLUTION

Virtual

Data Center

Cloud

Data Center

Physical

Data Center

BARRIER TO FULL CLOUD ADOPTION

CAN I VIRTUALIZE MY…

Compute Network Storage

7 MYTHS OF SDN

Truth:

It will fuel

hardware

innovation

Truth:

Considerable

intelligence

stays

decentralized

Truth:

OpenFlow is

just a protocol

and probably

not the most

important one

for SDN

Truth:

It will happen

step-by-step

Truth:

We will begin

to see the

impact in 2013

Truth:

It applies to

all networking

and networking

services

Truth:

Opex

reduction

is more

significant

SDN is a forklift upgrade

Truth:

It will fuel

hardware

innovation

Truth:

Considerable

intelligence

stays

decentralized

Truth:

OpenFlow is

just a protocol

and probably

not the most

important one

for SDN

Truth:

It will happen

step-by-step

Truth:

We will begin

to see the

impact in 2013

Truth:

It applies to

all networking

and networking

services

Truth:

Opex

reduction

is more

significant

Myth:

It’s only

about software

Myth:

It’s only about

centralization

Myth:

It’s only

about

OpenFlow

Myth:

It’s going

to happen

immediately

Myth:

It’s only about

datacenter

networking

Myth:

It’s only about

reducing

CAPEX

Myth:

It’s going to

take forever

7 MYTHS OF SDN

8th myth

Google: “7 Truths of SDN”

myth

8th

SDN is a forklift upgrade

myth

myth

8th

SDN is a forklift upgrade

myth

Truth:

SDN will work with your existing network

INITIAL APPROACH TO SDN

Separates the data plane from the control plane

Completely centralizes the control plane

First packet of every flow is punted to the controller - reactive

Uniform flat network

Very large forwarding table in switches

Tenant changes affects all switches in path

Replaces existing network and protocols

REACTIVE END-TO-END NETWORK

REACTIVE END-TO-END NETWORKS

First packet of every flow

is punted to controller

Controller reactively programs every flow

on every switch on path

Per-tenant state in physical network:

Switches contain many flows

High latency. Low scalability. Fragile. Fork-lift upgrade.

A TWO DIMENSIONAL SPECTRUM

REACTIVE

PROACTIVE

END-TO-END OVERLAY

Reactive

End-to-End

Aggregated

Flows in

Core

Proactive Flow

Installation

Proactive

Overlay

A BETTER APPROACH TO SDN

Underlay physical network provides industry standard L2 & L3 forwarding

Tenant state only at the network edge – server hypervisors & gateways

Controller proactively installs forwarding state

Much smaller forwarding table in switches

Tenant changes don’t affect physical network

Incremental evolution of existing network and protocols

PROACTIVE OVERLAY NETWORK

PROACTIVE OVERLAY NETWORKS

Packets are not

punted to controller

Existing protocols establish

IP fabric underlay

No per-tenant state in physical network:

Switches only know physical servers

Low latency. High scalability. Robust. Evolutionary.

Controller proactively programs virtual

overlay switches & gateways only

WHAT ABOUT OPENFLOW?

OpenFlow is a control protocol

OpenFlow 1.3 was initially used for Reactive End-to-End networks

Very low level – i.e. programming hardware TCAM entries

Not ideally suited to establish high-level, proactive overlays

That said, OpenFlow can be used for all SDN approaches:

Reactive End-to-End, Aggregated Flows in Core, Proactive Flow

Installation, & Proactive Overlay

Alternative control protocols better suited to the Proactive Overlay model

XMPP or OVS-DB

WANT MORE DETAIL?

Blog by Bruno Rijsman – Juniper SDN Architect

Part One:

http://forums.juniper.net/t5/The-New-Network/Applying-the-Goldilocks-

principle-to-SDN/ba-p/190245

Part Two:

http://forums.juniper.net/t5/The-New-Network/Myth-8-Software-Defined-

Networking-requires-a-forklift-upgrade/ba-p/190255

Whitepaper by Bruno Rijsman – Juniper SDN Architect

http://www.juniper.net/us/en/local/pdf/whitepapers/2000515-en.pdf

END-TO-END VS. OVERLAY

SDN Controller

Configuration Analytics

Control

Virtualized Server

VM VM VM

Virtualized Server

VM VM VM IP fabric (underlay network)

Horizontally

scalable

Highly available

Federated

SDN CONTROLLER

Control

Orchestrator

OVERLAY ARCHITECTURE

JUNOS-V CONTRAIL CONTROLLER

SDN Controller

Configuration Analytics

Control

Virtualized Server

VM VM VM

Virtualized Server

VM VM VM IP fabric (underlay network)

Juniper Qfabric/QFX/EX

or 3rd party underlay switches

Juniper MX

or 3rd party gateway routers

Tenant VMs

Horizontally

scalable

Highly available

Federated

JunosV Contrail Controller

KVM Hypervisor +

JunosV Contrail (L2 & L3)

SDN CONTROLLER

Control

Orchestrator

JUNOS-V CONTRAIL CONTROLLER

SDN Controller

Configuration Analytics

Control

Virtualized Server

VM VM VM

Virtualized Server

VM VM VM IP fabric (underlay network)

Juniper Qfabric/QFX/EX

or 3rd party underlay switches

Juniper MX

or 3rd party gateway routers

Tenant VMs

BGP

Federation

Horizontally

scalable

Highly available

Federated

BGP

Clustering

JunosV Contrail Controller

KVM Hypervisor +

JunosV Contrail (L2 & L3)

REST

XMPP

SDN CONTROLLER

Control

Orchestrator

XMPP BGP + Netconf

JUNOSV CONTRAIL SERVICE CHAINING

SDN Controller

Configuration Analytics

Control

Virtualized Server

VM VM VM

Virtualized Server

VM VM VM IP fabric (underlay network)

REST

SDN CONTROLLER

Control

Orchestrator

Virtualized Server

Virtual Service

Virtualized Server

Virtual Service

XMPP XMPP XMPP

Overlay establishes a serial, “chain”

through multiple virtualized services

STANDARDS

Overlay control plane protocols:

XMPP: RFC 6120, draft-marques-l3vpn-end-system

BGP L3VPN: RFC 4364

BGP EVPN: draft-ietf-l2vpn-evpn

NetConf: RFC 6241

Multicast: draft-marques-l3vpn-mcast-edge

Overlay data plane encapsulation:

MPLS over GRE: RFC 4797

VXLAN (encapsulation only): draft-mahalingam-dutt-dcops-vxlan

Underlay control plane protocols:

Existing layer-2 or layer-3 protocols

Overall architecture

IETF NVO3 WG

ETSI NFV ISG

Centralize Controller Centralize Management

1 LICENSING MODEL

JUNIPER SOFTWARE ADVANTAGE

Full Use/Elastic Transferable Software

Lifetime Assurance

Standard Protocols Centralize Separate

Networking Planes

Use the cloud Common Platform Apply Broadly

Across Domains

6 PRINCIPLES

4 STEP ROADMAP

Extract Services Optimize the Hardware

JUNIPER’S SDN STRATEGY: 6-4-1

Centralize Controller Centralize Management

1 LICENSING MODEL

JUNIPER SOFTWARE ADVANTAGE

Full Use/Elastic Transferable Software

Lifetime Assurance

Standard Protocols Centralize Separate

Networking Planes

Use the cloud Common Platform Apply Broadly

Across Domains

6 PRINCIPLES

4 STEP ROADMAP

Extract Services Optimize the Hardware

JUNIPER’S SDN STRATEGY: 6-4-1

Centralize Management

Centralize Controller Centralize Management

1 LICENSING MODEL

JUNIPER SOFTWARE ADVANTAGE

Full Use/Elastic Transferable Software

Lifetime Assurance

Standard Protocols Centralize Separate

Networking Planes

Use the cloud Common Platform Apply Broadly

Across Domains

6 PRINCIPLES

4 STEP ROADMAP

Extract Services Optimize the Hardware

JUNIPER’S SDN STRATEGY: 6-4-1

Extract Services

Centralize Controller Centralize Management

1 LICENSING MODEL

JUNIPER SOFTWARE ADVANTAGE

Full Use/Elastic Transferable Software

Lifetime Assurance

Standard Protocols Centralize Separate

Networking Planes

Use the cloud Common Platform Apply Broadly

Across Domains

6 PRINCIPLES

4 STEP ROADMAP

Extract Services Optimize the Hardware

JUNIPER’S SDN STRATEGY: 6-4-1

Centralize Controller

JUNIPER PARTNERS Making Clouds Happen