floodlight overview & performance comparison by patrick huang

33
Floodlight Overview & Performance Comparison Speaker: Patrick Huang Advisor: H.P. Wen

Upload: patrick-huang

Post on 08-Dec-2014

1.689 views

Category:

Technology


1 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Floodlight overview & performance comparison by patrick huang

Floodlight Overview & Performance Comparison

Speaker: Patrick HuangAdvisor: H.P. Wen

Page 2: Floodlight overview & performance comparison by patrick huang

Outline

0Flashlight Overview 0A Flexible OpenFlow-Controller Benchmark0Comparing OpenFlow Controller Paradigms

Scalability: Reactive and Proactive

Page 3: Floodlight overview & performance comparison by patrick huang

Outline

0Flashlight Overview 0Reference list:

0 Project Floodlight, http://www.projectfloodlight.org/floodlight/

0 Slideshare0Big Switch

0A Flexible OpenFlow-Controller Benchmark0Comparing OpenFlow Controller Paradigms

Scalability: Reactive and Proactive

Page 4: Floodlight overview & performance comparison by patrick huang

Why Floodlight?

0 Openflow 0 Work with phy- and vir- switches that speak openflow protocol

0 Apache Licensed0 Use floodlight for any purpose

0 Open community0 Developed by open community

0 Easy to use0 Floodlight is drop dead simple to build and run.

0 enterprise-class

Page 5: Floodlight overview & performance comparison by patrick huang

OpenFlow implemented by Floodlight0 Application Tier

0 Control Plane Tier0 Data plane Tier

0 Indigo Data Plane Interface0 an open source project 0 aimed at enabling support for

OpenFlow on physical and hypervisor switches.  

Page 6: Floodlight overview & performance comparison by patrick huang

OpenFlow implemented by Floodlight

0 Floodlight 0 An Openflow

controller ,and 0 A collection of applications

built on top the floodlight controller

0 Applications on top of it0 solve different user needs

over the network

Page 7: Floodlight overview & performance comparison by patrick huang

OpenFlow implemented by Floodlight0 Flood light controller

0 Core service of common interest to SDN applications

0 Module Application0 Applications with higher bandwidth communication with controller

0 REST Application0 Application in any language leveraging service via REST API exposed by

controller modules and module applicaitons

Page 8: Floodlight overview & performance comparison by patrick huang

Floodlight ArchitectureTopology - Tracks links between hosts and switches

Device Manager - Tracks devices in the network (MACs, IPs, etc.)

Storage - Abstraction layer for storing controller storage. Memory is used.

Counter Store – Openflow + Floodlight stats

Routing / Forwarding – Core engine for storing, calculating paths and installing flows.

Web UI- Rest APIs Topology

Device Manager

Learning Switch

Hub

Web

Storage

CounterStore

Floodlight Controller

Routing / Forward

Static FlowPusher

REST APIs

Module App

Page 9: Floodlight overview & performance comparison by patrick huang

Floodlight ArchitectureLearning Switch - Can replace Routing / Forwarding(v)

Hub - Can replace routing / forwarding(x)

Main difference: REST API provider

TopologyDevice

Manager

Learning Switch

Hub

Web

Storage

CounterStore

Floodlight Controller

Routing / Forward

Static FlowPusher

REST APIs

Module App

Page 10: Floodlight overview & performance comparison by patrick huang

11

Open ControllersName Lang Platform(

s)License Original

AuthorNotes

OpenFlow Reference

C Linux OpenFlow License

Stanford/Nicira

not designed for extensibility

NOX Python, C++

Linux GPL Nicira actively developed

Beacon Java Win, Mac, Linux, Android

GPL (core), FOSS Licenses for your code

David Erickson (Stanford)

runtime modular, web UI framework, regression test framework

Maestro Java Win, Mac, Linux

LGPL Zheng Cai (Rice)

Trema Ruby, C Linux GPL NEC includes emulator, regression test framework

Floodlight Java Win, Mac, Linux

Apache Big Switch Apache licensed, actively developed

Page 12: Floodlight overview & performance comparison by patrick huang

Outline

0Flashlight Overview 0A Flexible OpenFlow-Controller Benchmark

0 Soft Defined Networking(EWSDN), 2012 Euro Workshop on0 Univ. of Wurzburg, Wurzburg, Germany

0Comparing OpenFlow Controller Paradigms Scalability: Reactive and Proactive

Page 13: Floodlight overview & performance comparison by patrick huang

A Flexible OpenFlow-Controller Benchmark

0Goal0 Introduce a tool to achieve a flexible Openflow Controller

benchmark0Methods

0 The Benchmark creates a set of message-generating virtual switches

Page 14: Floodlight overview & performance comparison by patrick huang

Related work

0Measure OF switching performance0 OF switches not designed as flow switches

Often performance bottlenecks

0Cbench

Page 15: Floodlight overview & performance comparison by patrick huang

Result: Mean Round Trip Time

0 RTT: Interval of0 Packet-In message is dispatched from the virtual switch to the

controller 0 Packet-Out or FlowMod message is received by the switch.

Page 16: Floodlight overview & performance comparison by patrick huang

Result: RTT0 Flood light and Nox

Behave similarly0 Increase rapidly from

0 200ms, 1 switch0 6 seconds, 30

switches

0 Both controllers are under heavy load at the point 0 Due to weak

hardware

Page 17: Floodlight overview & performance comparison by patrick huang

Result: RTT0 Maestro start at RTT

6ms0 Larger the no. of

switches, larger RTT0 But, far steady than two

others

Page 18: Floodlight overview & performance comparison by patrick huang

Result: Send and ReceptionRate

0 The rate is accepts packets0 Provide insights into rate control and polling strategies

0No. of packets sent from switches to the controller

Page 19: Floodlight overview & performance comparison by patrick huang

Virtual Switch Packet-in Send-rate0 Send rate: From switches to

controller through OF secure channel

0 Floodlight0 do not increase obviously

with the no. of switches 0 Start at 10000 pps, one

switch0 Increase to 38000 pps

0 NOX0 10000-70000

0 Maestro0 5000-140000

Page 20: Floodlight overview & performance comparison by patrick huang

Virtual Switch Packet-in Send-rate0 Suggestion:

0 NOX and Floodlight0 Implementation of rate control

mechanism

0 Maestro0 Accept packet in a best effort

manner

Page 21: Floodlight overview & performance comparison by patrick huang

Virtual Switch Packet-out Reception-rate

0 The no. of responses the switches recieves

0 No increase for floodlight and NOX0 Stable at 10000 pps

0 Maestro0 From 5000-135000

0 Outstanding packets0 No. of unanserwed

messages by controller

Page 22: Floodlight overview & performance comparison by patrick huang

Virtual Switch Packet-out Reception-rate

0 Outstanding packets0 No. of unanserwed

messages by controller

0 140000-135000

Page 23: Floodlight overview & performance comparison by patrick huang

Outline

0Flashlight Overview 0A Flexible OpenFlow-Controller Benchmark0Comparing OpenFlow Controller Paradigms

Scalability: Reactive and Proactive0 Advanced Information Networking and Applications (AINA),

20130 Univ. Estadual do Ceara (UECE), Fortaleza, Brazil0 Citation: 1

Page 24: Floodlight overview & performance comparison by patrick huang

Comparing OpenFlow Controller Paradigms Scalability: Reactive and Proactive

0Analysis 0 Demo the flaw of reactive approach

0Conclusion0 Indicate the effectiveness of a hybrid approach to improve the

efficiency and scalability of OF architecture

Page 25: Floodlight overview & performance comparison by patrick huang

Introduction

0 Issue: Scalability 0Only one controller

0 As the no. of OpenFlow switches increases0Control messages to the centralized controller grows0With increase of network diameter, switches have longer setup

delay0Bounded by controller’s processor power

Page 26: Floodlight overview & performance comparison by patrick huang

Related Work

0Hyperfolw:0 To provide scalability

Use as many controllers as necessary But keep network control logically centralized

0DevoFlow, to provide scalability0 devolve network control to switch 0 Introduce two new mechanisms to be imple. On swtich

Rule cloningLocal actions

Page 27: Floodlight overview & performance comparison by patrick huang

Related Work• Source-Flow, to reduce no. of flow entries

– Try to reduce no. of flow entries• Use MPLS-like tunneling approach to reduce Ternary Content

Addressable Memory used space

Page 28: Floodlight overview & performance comparison by patrick huang

OF architecture

• Reactive– First packets of flow triggers controller to insert flow table – Pros

efficient use of flow table memory – Cons

Cause setup timeHard dependency, connection must retain

Page 29: Floodlight overview & performance comparison by patrick huang

OF architecture

• Proactive– Controller pre-populate flow table in switches– Pros

zero setup timeSoft dependency

– ConsHard management

Page 30: Floodlight overview & performance comparison by patrick huang

Evaluating OpenFlow controller’s Parafigm

Page 31: Floodlight overview & performance comparison by patrick huang

Evaluating OpenFlow controller’s Parafigm

• Run Cbench to stress the controller’s capacity• The benchmarking measurement – flows per sec that can be processed by controller

Page 32: Floodlight overview & performance comparison by patrick huang

Result: Real network&Mininet

Page 33: Floodlight overview & performance comparison by patrick huang

Thank You