Logically Centralized? State Distribution Trade-offs in Software Defined Networks.

Slides:



Advertisements
Similar presentations
SDN Controller Challenges
Advertisements

Software-defined networking: Change is hard Ratul Mahajan with Chi-Yao Hong, Rohan Gandhi, Xin Jin, Harry Liu, Vijay Gill, Srikanth Kandula, Mohan Nanduri,
Chapter 2 Exploring Data with Graphs and Numerical Summaries
Unit 1.1 Investigating Data 1. Frequency and Histograms CCSS: S.ID.1 Represent data with plots on the real number line (dot plots, histograms, and box.
Scalable Multi-Cache Simulation Using GPUs Michael Moeng Sangyeun Cho Rami Melhem University of Pittsburgh.
Maximum Battery Life Routing to Support Ubiquitous Mobile Computing in Wireless Ad Hoc Networks By C. K. Toh.
Jaringan Komputer Lanjut Packet Switching Network.
CSE 534 Fundamentals of Computer Networks Lecture 4: Bridging (From Hub to Switch by Way of Tree) Based on slides from D. Choffnes Northeastern U. Revised.
ElasticTree: Saving Energy in Data Center Networks Brandon Heller, Srini Seetharaman, Priya Mahadevan, Yiannis Yiakoumis, Puneed Sharma, Sujata Banerjee,
Toward Transitional SDN Deployment in Enterprise Networks Marco Canini with Dan Levin, Stefan Schmid, Anja Feldmann TU Berlin / Telekom Innovation Labs.
Data and Computer Communications Ninth Edition by William Stallings Chapter 12 – Routing in Switched Data Networks Data and Computer Communications, Ninth.
Making Cellular Networks Scalable and Flexible Li Erran Li Bell Labs, Alcatel-Lucent Joint work with collaborators at university of Michigan, Princeton,
OpenFlow-Based Server Load Balancing GoneWild
Towards Virtual Routers as a Service 6th GI/ITG KuVS Workshop on “Future Internet” November 22, 2010 Hannover Zdravko Bozakov.
Web Caching Schemes1 A Survey of Web Caching Schemes for the Internet Jia Wang.
©NEC Laboratories America 1 Hui Zhang Samrat Ganguly Sudeept Bhatnagar Rauf Izmailov NEC Labs America Abhishek Sharma University of Southern California.
Overview Distributed vs. decentralized Why distributed databases
Internet Networking Spring 2002 Tutorial 13 Web Caching Protocols ICP, CARP.
Spring Routing & Switching Umar Kalim Dept. of Communication Systems Engineering 06/04/2007.
Handout # 4: Scaling Controllers in SDN - HyperFlow
A Scalable, Commodity Data Center Network Architecture Mohammad Al-Fares, Alexander Loukissas, Amin Vahdat Presented by Gregory Peaker and Tyler Maclean.
Building a Strong Foundation for a Future Internet Jennifer Rexford ’91 Computer Science Department (and Electrical Engineering and the Center for IT Policy)
1 Logically Centralized? State Distribution Trade-offs in Software Defined Networks Written By Dan Levin, Andreas Wundsam, Brandon Heller, Nikhil Handigol,
A Scalable, Commodity Data Center Network Architecture.
PostgreSQL Enhancement PopSQL Daniel Basilio, Eril Berkok Julia Canella, Mark Fischer Misiu Godfrey, Andrew Heard.
Not All Microseconds are Equal: Fine-Grained Per-Flow Measurements with Reference Latency Interpolation Myungjin Lee †, Nick Duffield‡, Ramana Rao Kompella†
Virtual LANs. VLAN introduction VLANs logically segment switched networks based on the functions, project teams, or applications of the organization regardless.
UCSC 1 Aman ShaikhICNP 2003 An Efficient Algorithm for OSPF Subnet Aggregation ICNP 2003 Aman Shaikh Dongmei Wang, Guangzhi Li, Jennifer Yates, Charles.
Protocols and the TCP/IP Suite Chapter 4. Multilayer communication. A series of layers, each built upon the one below it. The purpose of each layer is.
Institute of Computer and Communication Network Engineering OFC/NFOEC, 6-10 March 2011, Los Angeles, CA Lessons Learned From Implementing a Path Computation.
Google File System Simulator Pratima Kolan Vinod Ramachandran.
Top-Down Network Design Chapter Nine Developing Network Management Strategies Oppenheimer.
Network Aware Resource Allocation in Distributed Clouds.
Wei Gao1 and Qinghua Li2 1The University of Tennessee, Knoxville
CS 3700 Networks and Distributed Systems Inter Domain Routing (It’s all about the Money) Revised 8/20/15.
Budapest University of Technology and Economics Department of Telecommunications and Media Informatics Optimized QoS Protection of Ethernet Trees Tibor.
Architectures and Algorithms for Future Wireless Local Area Networks  1 Chapter Architectures and Algorithms for Future Wireless Local Area.
BOX PLOTS (BOX AND WHISKERS). Boxplot A graph of a set of data obtained by drawing a horizontal line from the minimum to maximum values with quartiles.
Aaron Gember, Theophilus Benson, Aditya Akella University of Wisconsin-Madison.
11 CLUSTERING AND AVAILABILITY Chapter 11. Chapter 11: CLUSTERING AND AVAILABILITY2 OVERVIEW  Describe the clustering capabilities of Microsoft Windows.
Routing Networks and Protocols Prepared by: TGK First Prepared on: Last Modified on: Quality checked by: Copyright 2009 Asia Pacific Institute of Information.
CellSDN: Software-Defined Cellular Core networks Xin Jin Princeton University Joint work with Li Erran Li, Laurent Vanbever, and Jennifer Rexford.
Evolving Toward a Self-Managing Network Jennifer Rexford Princeton University
OPERATING SYSTEMS CS 3530 Summer 2014 Systems and Models Chapter 03.
Spring 2000CS 4611 Routing Outline Algorithms Scalability.
CSci8211: SDN Controller Design 1 Overview of SDN Controller Design  SDN Re-cap  SDN Controller Design: Case Studies  NOX Next Week:  ONIX  ONOS 
Network Virtualization Sandip Chakraborty. In routing table we keep both the next hop IP (gateway) as well as the default interface. Why do we require.
Coping with Link Failures in Centralized Control Plane Architecture Maulik Desai, Thyagarajan Nandagopal.
Preliminaries: EE807 Software-defined Networked Computing KyoungSoo Park Department of Electrical Engineering KAIST.
12.Nov.2007 Capacity of Ad Hoc Wireless Networks Jinyang Li Charles Blake Douglas S. J. De Coutu Hu Imm Lee Robert Morris Paper presentation by Tonio Gsell.
Software Defined Networking BY RAVI NAMBOORI. Overview  Origins of SDN.  What is SDN ?  Original Definition of SDN.  What = Why We need SDN ?  Conclusion.
Yiting Xia, T. S. Eugene Ng Rice University
Xin Li, Chen Qian University of Kentucky
SDN challenges Deployment challenges
OPERATING SYSTEMS CS 3502 Fall 2017
The DPIaaS Controller Prototype
Chris Cai, Shayan Saeed, Indranil Gupta, Roy Campbell, Franck Le
Hydra: Leveraging Functional Slicing for Efficient Distributed SDN Controllers Yiyang Chang, Ashkan Rezaei, Balajee Vamanan, Jahangir Hasan, Sanjay Rao.
ETHANE: TAKING CONTROL OF THE ENTERPRISE
Chapter 5 : Describing Distributions Numerically I
Overview of SDN Controller Design
Software Defined Networking (SDN)
Intra-Domain Routing Jacob Strauss September 14, 2006.
Software Defined Networking (SDN)
ElasticTree: Saving Energy in Data Center Networks
   Storage Space Allocation at Marine Container Terminals Using Ant-based Control by Omor Sharif and Nathan Huynh Session 677: Innovations in intermodal.
Cloud Computing Architecture
2019/5/13 A Weighted ECMP Load Balancing Scheme for Data Centers Using P4 Switches Presenter:Hung-Yen Wang Authors:Peng Wang, George Trimponias, Hong Xu,
2019/10/9 A Weighted ECMP Load Balancing Scheme for Data Centers Using P4 Switches Presenter:Hung-Yen Wang Authors:Jin-Li Ye, Yu-Huang Chu, Chien Chen.
Control-Data Plane Separation
Presentation transcript:

Logically Centralized? State Distribution Trade-offs in Software Defined Networks

Presentation Overview Overview Problem Description Key Tradeoffs Evaluation Summary References

Problem Physically centralized control is inadequate, it limits o responsiveness o reliability o scalability Need a physically distributed control plane on which logically centralized control plan operates o Causes tradeoffs (solution to problem becomes new problem!)  Consistency models Strongly and Eventually Consistent  how much will performance suffer? Categorize tradeoffs, simulate and measure

Questions How centralized or distributed should the network control plane be? How does distributed SDN state impact the performance of a logically centralized control application? When the underlying distributed control plane state leads to inconsistency or staleness in global network view, how much does network performance suffer?

Key State Exchange Points in SDN Dotted lines are State Exchange Points in SDN Physical Devices are grouped into domains o Ctrl 1, Ctrl 2 State flows up, control flows down NIB is updated from Physical as well as other controllers

Key Tradeoffs Staleness vs Optimality o Cost to achieve consistent state in global network view entails higher rates of control synchronization and communication Application Logic Complexity vs Robustness to Inconsistency o Example of end-to-end argument

Tradeoff #1 - Staleness vs Optimality Each controller maintains a NIB o global view of network NIB is updated by o other controllers o physical network control application performance suffers in presence of inconsistency consistent state requires higher rates of control communication

Tradeoff #2 App Complexity vs Robustness Applications can either be aware of state inconsistencies or not If they are aware, complexity of application increases o more to be aware of If they are not aware, performance suffers o take action on stale data

Evaluation - Example Application Load Balancing o Objective is to minimize the maximum link utilization in network Two Implementations o Link Balancer Controller (LBC) o Separate State Link Balancer Controller (SSLBC)

Example Application - LBC Link Balancer Controller (LBC) o Simpler of two applications o Inspired by “Load Balancing Gone Wild” o Application is presented with a global network view when certain events are triggered, application uses this view to calculate paths, the path with lowest max link utilization is chosen to assign next arriving flow and forwarding state is installed in network

Example Application - SSLBC Separate State Link Balancer Controller (SSLBC) o Application keeps fresh intra-domain physical network state from updates learned through inter- domain controller synchronization events o Potentially stale global network view is used o global network view guides each application instance to redistribute scaled fraction of its local link imbalance on a flow-by-flow arrival basis

Evaluation - Network Architecture Two co-operating controller domains Each domain has single switch Upstream traffic considered negligible Table describes link capacities servers are identical replicas fat interdomain links Load Balancer objective o minimize Root Mean Squared Error (RMSE)

Evaluation - Workloads Deterministic, controlled workload o Flow arrival rate driven by sin function More Realistic Workload o exponentially distributed flow inter-arrival times o modulate mean of exponential distribution by wave function Common o vary time interval between NOS synchronization events

Evaluation Results - How to Read the Graph 32 arrivals per timestep, fixed flow duration of 2 timesteps, total of 64 flows active Vertical line represents sync occurring lower % is better Top graph corresponds to 16 in bottom graph NOS Sync Period is how often state synched Box shows the center half of data with median marked whiskers show 95 percentiles outliers are marked 0 corresponds to balanced loads

Evaluation Results - LBC, SSLBC, Simple Workload As global network view becomes inconsistent, application performance can suffer (left graphs) awareness of underlying distributed state leads to less sensitivity to NOS staleness (right graphs) SSLBC may not perform as well with high rates of NOS synchronization as it is more conservative (0,1 in both graphs)

Evaluation Results - LBC, SSLBC, Realistic Workload SSLBC is still better!

Evaluation Summary Results o View staleness significantly impacts optimality  As global network view becomes inconsistent, application performance can suffer o Application robustness to inconsistency increases when application logic is aware of distribution Other Applications - not covered o Distributed firewalls, IDS, admission enforcement, routing, middle-box application

Questions How centralized or distributed should the network control plane be? How does distributed SDN state impact the performance of a logically centralized control application? When the underlying distributed control plane state leads to inconsistency or staleness in global network view, how much does network performance suffer?

References Logically Centralized? State Distribution Trade-offs in Software Defined Networks; Dan Levin, Andreas Wundsam, Brandon Heller, Nikhil Handigol, Anja Feldmann