Andreas Kassler OpenFlow and Wireless Mesh Peter Dely ......Zukunft der Netze 2012, Wien, Austria, September 28th 2012 OpenFlow and Wireless Mesh Networks – Andreas Kassler slide

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012

OpenFlow and Wireless Mesh Networks – Andreas Kassler slide 1

OpenFlowand

Wireless Mesh Networks

Andreas Kassler

Peter Dely

Karlstads Universitet

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Outline

Wireless Mesh Networks and SDN Use case 1: Joint Rate Control and Routing with OpenFlow Use case 2: Towards Cloud based WLAN MAC processing Use case 3: Fast WLAN HOs Summary and conclusion

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Mesh Networks and the Future Wireless Internet

Wireless Mesh Networks– Broadband for all

• Urban Mesh Networks• Developing regions

– Single hop vs. multihop– Part of future wireless

internet– Deployed in ISM bands

Source: Tropos

Source: TroposSource: Tropos

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Large Deployments (e.g. Enterprise WLANs)

Services

MN

WLANManagement

System

WLANController

CentralizedManagement

Mobility Mgmt.Loadbalancing

ChannelAssignment

AP

Limited coverage

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Large Deployments for Mesh Networks

Services

MAP

MN

MeshManagement

System

MeshController

Centralized vs. DistributedManagement

Mobility Mgmt.Loadbalancing

ChannelAssignmentRouting

MGW

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Vendor Specific Solutions??

Services

MAP

MeshManagement

System

MeshController

MGW

How to avoid Vendor lock-in?

How to open up innovation?

Flexibility by using open interfaces?

SDN

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


SDN and OpenFlow

Application Layer

Infrastructure Layer

Control Layer

Business ApplicationsBusiness ApplicationsBusiness Applications

Network Device Network Device Network Device

Network Device Network Device

SDN Control Software

Business ApplicationsBusiness ApplicationsNetworking Service

API API API

Control Data Plane Interface (e.g. OpenFlow)

Enterprise WLANWireless Mesh

e.g. Switch or AP

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Outline


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Problem Description: Beer vs. Internet Connectivity?

Stupid Internet, too slow again! I need more

beer!

Example: ”Hotmesh” Project with Telekom Innovation Laboratories

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Problem Description: Technical View

Example: ”Hotmesh” Project with Deutsche Telekom Laboratories

Default policy:– STA associates to MAP with

highest RSSI (signal strength)– MAP uses some routing

metric to route traffic to MGW– All STA associated to one

MAP use same MGW

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Problem Description


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Problem Description


OverloadOverload

Low loadLow loadLow loadLow load

OverloadOverload

OverloadOverload

Problems:– Client association

does not consider load on MAP and capacity of path to Internet (Wireless and Wired)

– Optimum MAP might change over time

– If a client moves between two MAPs the IP must be maintained to preserve end-to-end connectivity

Solution (classical)– Modify routing protocol– Using tunnels

Want to control MAP/GW Selection and Routing

INDEPENDENT ofRouting protocol used

Mathematical Problem is a MILP

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


OpenFlow for Wireless Mesh Networks

Services

MAP

MeshManagement

System

NOX

MGW

RoutingRate Control

User Association

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Services

MAP

MeshManagement

System

NOX

MGW

RoutingRate Control

User Association

OpenFlow gives us:Simple programmingCentral network view

Can use standard optimizers (CPLEX)

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Services

MeshManagement

System

NOX

RoutingRate Control

User Association

Matching Problem

Flow Maximization

Routing

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Services

MeshManagement

System

NOX

RoutingRate Control

User Association

1. minrate = minrate_old = 02. alpha = 0%3. Find initial STA/MAP matching (e.g. each STA

associates to MAP with min-hop count to GW)4. Calculate max-flow, so that each no two Xs, Xt differ

by more than alpha (Xs = xv/(#STA associated at v))5. minrate = min(X1,…,Xs)6. alpha = inf7. Calculate max-flow, so that each MAP v gets at least

minrate *(#STA associated at v)8. Reshuffle STA between MAP, so that minrate increases9. Calculate max-flow with new matching, ensuring at

least minrate to each STA10. If minrate > minrate_old:

– minrate_old = minrate– GOTO 4

11. Calculate routes

www.kau.se/en/kaumeshhttp://www.youtube.com/kaumesh

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


OpenFlow for Wireless Mesh Networks - Evaluation

Higher gateway speeds – Better gain

Fair Optimization of Mesh-Connected WLAN Hotspots, Peter Dely, Fabio D’Andreagiovanni, Andreas Kassler, under review at: Wiley Journal on Wireless Communications and Mobile Computing

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


OpenFlow for Wireless Mesh Networks - Evaluation

25 STA 50 STA 100 STA

Works better with more STA

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Outline


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Usecase 2: 802.11 MAC layer processing in Cloud

How can we offload some processing to data centers?

How can we deploy network applications in a vendor independent way?

How can we exploit the fast packet processing in hardware switches to control WLAN transmissions?

1. AP hardware and software are getting fatter and fatter

2. No standard, vendor-independent way to deploy network applications

3. Fast IEEE 802.11 PHY layers make centralized control planes difficult to implement

Example: ”EasyMobile” Project with Telekom Innovation Laboratories

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


CloudMAC Architecture

Virt

ual

WLA

N N

ICV

irtua

l W

LAN

NIC

Virtu

al

WLA

N N

IC

CloudMAC – Towards Software Defined WLANs, Jonathan Vestin, Peter Dely, Andreas Kassler, Nico Bayer, Hans Einsiedler, Christoph Peylo, in: Proceedings of MobiCom’12: The 18th Annual International Conference on Mobile Computing and Networking, Istanbul, August 2012. Won the 1st price in the ACM Mobicom Student Research Competition!

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


CloudMAC: TXPath

Virt

ual

WLA

N N

IC

DataIPDataIP802.11Ctrl

Lookup flow tableModify control header Output frame

DataIP802.11

Add IEEE 802.11 MAC header Add control header (Encrypt frame)

Read and remove control header IEEE 802.11 real-time MAC PHY processing Transmit frame

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


CloudMAC: Performance

CLOUDMACs additional delay due to the additional processing in software is limited!

Performance loss due to the header overhead and rule processing is tolerable!

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


CloudMAC Demo: Seamless AP Switching

Move traffic from WTP1 to WTP2 Mobility management or energy saving Standard IEEE 802.11: scan and re-

association CloudMAC: Association state in VAP no

reassociation

Move traffic

Change flow table entry

Median reduction in packet lossfrom approx. 10000 to 4

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Outline


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


UseCase 3: Fast HO or The Rose-Garden Problem

How can Obama walk from the garden to the oval office

while talking to Merkel without video freeze?

Example: ”Easymobile” Project with Telekom Innovation Laboratories/Stanford

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


A Technical Perspective

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Associate


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Time

BufferLevel

Quality forcurrent AP

starts to decrease

Connection completely

breaks

Video playout freezes

Scan for new APs Connect to

new AP

Video playout

resumes


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


BufferLevel

Video freeze time

Time

1. Start playout

already here

2. Initiate handovers

earlier

3. Scan for new APs before

connection breaks

4. Increase handover

speed


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


System Overview

When to initiate a handover?– Lawrence Chow, Bradley Collins, Nick Bambos, Nico Bayer, Hans Einsiedler, Christoph Peylo, Peter Dely, Andreas Kassler, “Playout-Buffer

Aware Hand-Off Control for Wireless Video Streaming”, Proceedings of Globecom 2012. What is a good AP to select? How can we aggregate BW from multiple APs

– Peter Dely, Andreas Kassler, Lawrence Chow, Nick Bambos, Nico Bayer, Hans Einsiedler, Christoph Peylo, “BEST-AP: Non-intrusive Estimation of Available Bandwidth and its Application for Dynamic Access Point Selection”, submitted to Elsevier Computer Communications Journal.

When to resume playing?– Lawrence Chow, Bradley Collins, Nick Bambos, Christoph Peylo, Hans Einsiedler, Nico Bayer, Peter Dely, Andreas Kassler, “Channel

Aware Rebuffering for Media Streaming with Handoff Control”, Submitted for publication in ICC 2013. How to increase the speed of handovers? How to scan for new APs?

– Peter Dely, Andreas Kassler, Lawrence Chow, Nick Bambos, Nico Bayer, Hans Einsiedler, Christoph Peylo, “BEST-AP: Non-intrusive Estimation of Available Bandwidth and its Application for Dynamic Access Point Selection”, Submitted for publication in Elsevier Computer Communications Journal, 2012.

– Peter Dely, Andreas Kassler, Lawrence Chow, Nick Bambos, Nico Bayer, Hans Einsiedler, Christoph Peylo, Daniel Mellado, Miguel Sanghez “Enabling Fast WLAN Handovers for Real-Time Video with Software Defined Networks”, Submitted for publication in The 1st International Workshop on High Mobility Wireless Communications, 2012.

Optimized HO scheduler BW estimation for each AP Preauthentication via APs to all

other APs in range

Constantly evaluate channel load and link quality for each client Report bandwidth estimations to

controller Fuses bandwidth estimations Informs clients about changes Distributes Pre-Authentication messages

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Station Architecture

Decides (based on buffer level, QoS..) when to scan for new APs when to initiate HO

Creates Control Messages to network Control Messages for lower layer

Interface for higher layers with unique IP/MAC-@

directs traffic to virtual WLAN cards, rewrites source MAC-@ Each virtual card can be

associated to a different AP on a different channel Only one physical card required

to scan and to be connected toseveral APs

Can be easily extended for Multi-Radio Platforms (e.g.

WLAN/LTE)

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Optimized HO from AP1 to AP2

Station AP2 Switch ControllerAP1

Handover Initiate

Handover Initiate

Pause Transmission

ACK

Update Forwarding Table

Handover Complete

Handover Complete

Resume Transmission

“Measurement and Analysis of Handover Latencies in IEEE 802.11i Secured Networks” by Martinovic et. all reports handover delays of 6-10 seconds for normal WLANs

Pre-association and Pre-authentication

make HOs fast

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Usecase 3: Fast HO - Evaluation

Mode

Freeze time

fraction

Number of freeze events

Maximum freeze

time (sec)Mean freeze time (sec)

Proposal/No dedicated scan card (scan every 2 sec) 0.73% 15 2.24 0.46

Proposal/Dedicated scan card (0.5 sec scan interval) 3.25% 11 6.24 2.24

Linux/Optimized scan (ch 36,40,44,48) 12.48% 55 8.43 2.01Linux/Standard scan (all channels) 28.30% 79 32.06 2.75

Each HO comes at a cost due to video freezing probability!

Optimization of WLAN associations considering handover costs, Peter Dely, Andreas Kassler, Nico Bayer, Hans Einsiedler and Christoph Peylo, accepted

for: EURASIP Journal on Wireless Communications and Networking

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Outline


Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Feature Benefit/DrawbackFlow based routing Distribute traffic over network on per node basis

Multi-path routing Can reduce outage duration during handover

Integrated traffic shaping Allows to enforce QoS settingsConcurrent use of standard routing protocols (e.g. OLSR)

Enables migration path for deployed networks

NOX provides centralized point of view

Allows to use standard optimization tools and implement fast new services

Webservice interface of NOX Easy integration with other services (e.g. A4C)

OpenVSwitch Enables clean kernel programming of end nodes with minimum effort

Virtualisation Enables Distributed MAC processing with Cloud Integration

Summary and Conclusion

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Thank [email protected]

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Performance Evaluation – Forwarding Speed

User space forwarding might impose some

performance bottlenecks on lowspeed devices!Memory speed issues!

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Performance Evaluation - Overhead

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Performance Evaluation – Flow activation time

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Performance Evaluation

Algorithm (Simulation)

Mesh Router (no AP IF)

Mesh GW (with AP IF)

Mesh Access Point

ns-2 simulations of Chaska topology Dual-radio nodes

– IEEE 802.11g for client access– IEEE 802.11a for backhaul– Autorate

Evaluation metric: 90%ile throughput Comparison

– SNR-based association (and min hop-count routing): default policy

– GW hop-count metric: STA associates to MAP with min hop-count to next GW

– GW min-cost metric (cost = 1/Link Speed)– HOTMESH/HC: Optimization algorithm, initial association is based on min hop-

count metric)– HOTMESH/MC: Optimization algorithm, initial association is based on min cost

metric

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Performance Evaluation (MultiRAT)

Algorithm Performance – Testbed Setup

18 Mbit/s PHY rate 10 STA

– Some STA (e.g. STA 1) can associate tomultiple MAP

7 mesh nodes– 3 mesh gateways (MGW) with

MAP interface– 2 mesh access points (MAP)– 2 mesh routers (MR)

Evaluation with different GW speeds TCP download from the “Internet” Multi-radio system Manual assignment of orthogonal

channels

Legend

A

B

C

D

G

E

F

1

23

4

10

5

6

7

8

9

MAP/MGW

MAP

MR

Backbone Link

Association Opp.

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Average Throughput

0

200

400

600

800

1000

1200

1400

1600

3Mbit/s 6Mbit/s

GW Speed

Thr

ough

put (

kbit/

s)SNR-basedHop-CountHotmesh/MC

25%

48%

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012



Throughput per STA, 3 Mbit/s GW Speed

0100200300400500600700800900

1000

1 2 3 4 5 6 7 8 9 10

Station ID

Thr

ough

put

(kbi

t/s)

SNR-basedHop-CountHotmesh/MC

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


s1

a bs2

s3

v1

v’1,1

v’1,2

v2

v’2,1

v’2,2

1

1

1

1

11

1

1

1

1

1

1

1

1

1

Mathematical Formulation – STA/MAP Matching

Each MAP can only serve a max number of number STA (“service slots”) Number of service slots depends on flow from Internet to MAP Maximum cardinality bipartite matching Generate a graph with service slots v’ and STA Calculate max-flow between ab If edge between s and v’ has weight 1 STA s

associates to MAP v

Can be solved using a LP

Alternatively: Hopcroft-Karp algorithm

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Mathematical Formulation – Flow Maximization

Maximize the flow from Internet to MAP Modeled as single-commodity max-flow problem Use connectivity graph and conflict graph Max-Flow tt’ can be calculated using an LP (or by modifying Ford–Fulkerson

algorithm)

Zuku

nftd

erNe

tze20

12, W

ien, A

ustri

a, Se

ptem

ber 2

8th

2012


Split flow here0.66 Mbit/s

0.33 Mbit/s

Mathematical Formulation – Routing

Find a path from t to each STA Paths to STA s need to be able to carry Xs flow Problem: Max-flow might not be achieved using unsplittable flows!

Documents

Andreas Kassler OpenFlow and Wireless Mesh Peter Dely ......Zukunft der Netze 2012, Wien, Austria, September 28th 2012 OpenFlow and Wireless Mesh Networks – Andreas Kassler slide