software-deﬁned data plane enhancing sdn and...

Software-Defined Data Plane Enhancing SDN and NFV

Aki Nakao The University of Tokyo

GSC 18 2014/7/22

1

2

Control-Plane Elements

Network Applications

Data-Plane Elements

Applications (NFV)

Control Plane

Data Plane

SDN Architecture

North Bound Interface (NBI)

South Bound Interface (SBI)

3



Data-Plane Elements

Applications (NFV)

Control Plane

Data Plane

Application Driven Thinking

Current SDN (bottom up)

Future SDN should be (top down)

4

Application-Driven Thinking Premise： Programmable networking has been enabled by SDN and NFV App-Driven Thinking: •  Think of “(killer) applications” first and then design network functions and interfaces APIs for SDN and NFV

•  Not just OPEX/CAPEX reduction but create new values via SDN and NFV

Lets start with an example application!

“Next-Gen MVNO”

5

An ISP needs app specific traffic control

Standard Mode (flat rate)

High Speed Mode (pay as you go)

bandwidth

150kbps

Bandwidth Control According to Apps !

6

MVNO Customers Need More Flexible Subscription

App-specific traffic control enables more fined grained subscription plans that can get an MVNO out of the “low cost” competition

Application Driven SDN

•  The current SDN is targeted at network operators (researchers) and not at application developers

•  Flow abstraction in Southbound Interface is for operators <Flow Pattern> <Action> <Stat>

•  App/Device abstraction is useful and intuitive for application developers

<App/Device> <Action><Stat>

Solution: Application-Driven Software Defined Networking 7

There is a gap between application developers and SDN’s programming interface

GENI Engineering Conference 20 Demo: App-‐Driven SDN QoS

Smartphone FLARE (Data Plane Programmable)

Chrome : Pass Thru Firefox: Block YouTube: Rate Limit

App Packet

8

Smartphones

Packet Marking

The Internet FLARE Network (SDN/NFV Enabled)

Header Payload

Add app/device information to packet trailers

Trailer

App/Device InformaTon

Smartphones (wearables)

Header Payload

Parse and remove “trailers” and map between flows and apps

FLARE（Deeply Programmable Node)

Header Payload

Traffic Engineering based on headers

SDN Controller

NTT Docomo

Our Proposal

SDN Controller

9

Gateway

MVNO Backhaul

Packet

Parse and Remove Trailers

Smartphones aWach app/device informaTon to packets FLARE detects app/device informaTon and creates mapping between flows and apps/devices

Freq

uency (Accesses/sec)

10

Demo

11

Demo

12

We won the best demo award! GEC20@UC Davis

13

Benefits •  Application Specific Traffic Engineering for MVNO •  Application Name Based •  Application Process Based（Fore/Background) •  Device Type Based •  Device State Based (Context / Location Aware) •  Parental Control •  Not by apps on devices, but by networking •  Additional Value-Add services for specific applications •  Differentiation for competing apps (e.g., Chrome vs. Firefox)

14

Contributions

•  Insert application information in trailers of packets (e.g., TCP SYN) •  Extensible to other protocols than TCP •  Similar to “Google Fast TCP Open” but for different purpose •  More bits usable than TCP/IP options •  Determine applications with 100% Accuracy •  Cooperation between end-systems and programmable nodes •  Extensible to supervised learning without app •  Machine learning using sampled data with app

Software Defined Data Plane

15



Programmable Data-Plane Elements

Applications

Control Plane

Data Plane

North-Bound Interface (NBI)

E.g. OpenFlow Switches

Packet Process

Packet Process

Publish API

Accelerate SDN Standardization

16

Network Applications Data-Plane

Elements

Southbound Interface

Application Driven Thinking

Operation and Evaluation Feedback -> Accelerate Standardization

Example

17

Sliceable Software Defined Data Planes

18




Applications

Control Plane

Data Plane



Packet Process

Packet Process

Publish API

Programmable Data-Plane Elements Programmable Data-Plane Elements Programmable Data-Plane Elements

19




Applications

Control Plane

Data Plane



Packet Process

Packet Process

Programmable Data-Plane Elements Programmable Data-Plane Elements Programmable Data-Plane Elements

SDN for Network Control NFV for data processing

Programmable Data-Plane Elements Programmable Data-Plane Elements Programmable Data-Plane Elements Programmable Data-Plane Elements

SDN data plane and NFV could be unified


Orchestrator

(Unifying SDN Software Defined Data Plane and NFV)

20

Physical Resources: CPU + NPU

CPU NPU(s) Data Bus

NIF

White Box

Intel x86 Tilera PCIe Gen3

FLARE

FLARE

21

Physical Resources: CPU + NPU (+ GPU)

CPU NPU(s) Data Bus

NIF

GPU Data Bus

Intel Phi Intel x86 Tilera PCIe Gen3

PCIe Gen3

22

GPU CPU NPU(s) Data Bus

Data Bus

NIF

Physical Resources

Slices (Virtual Resource Containers)

Virtualization ( Resource Container ) -> Slices of Resources

Sliceable !

FLARE

23


Data Bus

NIF

Physical Resources


FLARE

24


Data Bus

NIF

Physical Resources


SDN Sliceable Programmable D-plane

NFV Sliceable Programmable D-plane

SDN Programmable Control Plane

or

FLARE

FLARE Node Implementation

36-72 cores (upto 100-200 cores in future)

(board designed by NakaoLab)

x86 Processor

Many Core Processor

The University of Tokyo Confidential

Hierarchical Resource Management • General Purpose Processor(s) • Network Processor(s) • ...and more types of processors

25

FLARE-EX Prototype (32x10Gbps)

26

27

Toy-Block Networking

ToyBlock Networking GUI

Standardization Possibility

29

•  Deep (Data Plane) Programmability •  Slicing (Virtualization) •  Evolvable APIs •  Viewed as extension to SDN •  Also viewed as extension to NFV •  Rethinking architecture of SDN/NFV for enabling “Deep programmability”

•  Application Driven Thinking •  Top-Down approach is necessity

•  Toy Block Networking •  Accommodate a wide range of programmers •  Marketing of reusable network function “Toy Blocks”

“Software Eats Everything”

“Application Driven Thinking empowered with

Software Defined Data Plane will extend SDN/NFV further and accelerate standardization!”

30

Conclusion

software-deﬁned data plane enhancing sdn and...

Documents