An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.

Slides:



Advertisements
Similar presentations
1 Building a Fast, Virtualized Data Plane with Programmable Hardware Bilal Anwer Nick Feamster.
Advertisements

Monitoring very high speed links Gianluca Iannaccone Sprint ATL joint work with: Christophe Diot – Sprint ATL Ian Graham – University of Waikato Nick McKeown.
1.  Congestion Control Congestion Control  Factors that Cause Congestion Factors that Cause Congestion  Congestion Control vs Flow Control Congestion.
Improving TCP Performance over Mobile Ad Hoc Networks by Exploiting Cross- Layer Information Awareness Xin Yu Department Of Computer Science New York University,
SKELETON BASED PERFORMANCE PREDICTION ON SHARED NETWORKS Sukhdeep Sodhi Microsoft Corp Jaspal Subhlok University of Houston.
Geoff Salmon, Monia Ghobadi, Yashar Ganjali, Martin Labrecque, J. Gregory Steffan University of Toronto.
Top-Down Network Design Chapter Four Characterizing Network Traffic Copyright 2010 Cisco Press & Priscilla Oppenheimer.
Traffic Management - OpenFlow Switch on the NetFPGA platform Chun-Jen Chung( ) SriramGopinath( )
Lect-2: BasicsComputer Networks : An Introduction to Computer Networks Handout #2: Foundations and Basic Concepts Additional Reading Text book:
4-1 Network layer r transport segment from sending to receiving host r on sending side encapsulates segments into datagrams r on rcving side, delivers.
Chapter 4 Network Layer slides are modified from J. Kurose & K. Ross CPE 400 / 600 Computer Communication Networks Lecture 14.
10 - Network Layer. Network layer r transport segment from sending to receiving host r on sending side encapsulates segments into datagrams r on rcving.
1 Emulating AQM from End Hosts Presenters: Syed Zaidi Ivor Rodrigues.
Service Providers & Data Link & Physical layers Week 4 Lecture 1.
Review on Networking Technologies Linda Wu (CMPT )
IP-UDP-RTP Computer Networking (In Chap 3, 4, 7) 건국대학교 인터넷미디어공학부 임 창 훈.
Can Google Route? Building a High-Speed Switch from Commodity Hardware Guido Appenzeller, Matthew Holliman Q2/2002.
Comparing the Accuracy of Network Simulators for Packet-Level Analysis using a Network Testbed Chaudhry Usman Ali UNB, Fredericton.
Network Simulation Internet Technologies and Applications.
Localized Asynchronous Packet Scheduling for Buffered Crossbar Switches Deng Pan and Yuanyuan Yang State University of New York Stony Brook.
Chapter 4 Queuing, Datagrams, and Addressing
Practical TDMA for Datacenter Ethernet
IETF 90: VNF PERFORMANCE BENCHMARKING METHODOLOGY Contributors: Sarah Muhammad Durrani: Mike Chen:
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.
EstiNet Network Simulator & Emulator 2014/06/ 尉遲仲涵.
Traffic Modeling.
Wave Relay System and General Project Details. Wave Relay System Provides seamless multi-hop connectivity Operates at layer 2 of networking stack Seamless.
Fraunhofer FOKUSCompetence Center NET T. Zseby, CC NET1 IPFIX – IP Flow Information Export Overview Tanja Zseby Fraunhofer FOKUS, Network Research.
Tiziana Ferrari Diffserv deployment in the wide area: network design and testing1 Diffserv deployment in the wide area: network design and testing Tiziana.
LiNK: An Operating System Architecture for Network Processors Steve Muir, Jonathan Smith Princeton University, University of Pennsylvania
Traffic Management - OpenFlow Switch on the NetFPGA platform Chun-Jen Chung( ) Sriram Gopinath( )
Vladimír Smotlacha CESNET Full Packet Monitoring Sensors: Hardware and Software Challenges.
ECE 526 – Network Processing Systems Design Networking: protocols and packet format Chapter 3: D. E. Comer Fall 2008.
Salim Hariri HPDC Laboratory Enhanced General Switch Management Protocol Salim Hariri Department of Electrical and Computer.
Measurement and Modeling of Packet Loss in the Internet Maya Yajnik.
The Transport Layer application transport network data link physical application transport network data link physical application transport network data.
Requirements for Simulation and Modeling Tools Sally Floyd NSF Workshop August 2005.
Resource Mapping and Scheduling for Heterogeneous Network Processor Systems Liang Yang, Tushar Gohad, Pavel Ghosh, Devesh Sinha, Arunabha Sen and Andrea.
Multiplicative Wavelet Traffic Model and pathChirp: Efficient Available Bandwidth Estimation Vinay Ribeiro.
Vladimír Smotlacha CESNET High-speed Programmable Monitoring Adapter.
Lecture 12: Reconfigurable Systems II October 20, 2004 ECE 697F Reconfigurable Computing Lecture 12 Reconfigurable Systems II: Exploring Programmable Systems.
An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.
Simics: A Full System Simulation Platform Synopsis by Jen Miller 19 March 2004.
Modeling and Simulating Time- Sensitive Networking Harri Laine.
Kraków4FutureDaQ Institute of Physics & Nowoczesna Elektronika P.Salabura,A.Misiak,S.Kistryn,R.Tębacz,K.Korcyl & M.Kajetanowicz Discrete event simulations.
An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.
Efficient Gigabit Ethernet Switch Models for Large-Scale Simulation Dong (Kevin) Jin David Nicol Matthew Caesar University of Illinois.
1 CSE 5346 Spring Network Simulator Project.
An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin.
Top-Down Network Design Chapter Four Characterizing Network Traffic Copyright 2004 Cisco Press & Priscilla Oppenheimer.
Precision Measurements with the EVERGROW Traffic Observatory Péter Hága István Csabai.
Research Unit for Integrated Sensor Systems and Oregano Systems Cern Timing Workshop 2008 Patrick Loschmidt, Georg Gaderer, and Nikolaus Kerö.
Achievable Service Differentiation with Token Bucket Marking for TCP S. Sahu, D.Towsley University of Massachusetts P. Nain INRIA C. Diot Sprint Labs V.
Studies of LHCb Trigger Readout Network Design Karol Hennessy University College Dublin Karol Hennessy University College Dublin.
Client-server communication Prof. Wenwen Li School of Geographical Sciences and Urban Planning 5644 Coor Hall
2003 컴퓨터 통신 1 Foundation # 컴퓨터 통신 2 Outline A Detailed FTP Example Layering Packet Switching and Circuit Switching Some terms –Data rate, “Bandwidth”
PATH DIVERSITY WITH FORWARD ERROR CORRECTION SYSTEM FOR PACKET SWITCHED NETWORKS Thinh Nguyen and Avideh Zakhor IEEE INFOCOM 2003.
1 Scalability and Accuracy in a Large-Scale Network Emulator Nov. 12, 2003 Byung-Gon Chun.
Network Processing Systems Design
Ethernet Packet Filtering – Part 2 Øyvind Holmeide 10/28/2014 by.
Graciela Perera Department of Computer Science and Information Systems Slide 1 of 18 INTRODUCTION NETWORKING CONCEPTS AND ADMINISTRATION CSIS 3723 Graciela.
Chapter Objectives In this chapter, you will learn:
Architecture and Algorithms for an IEEE 802
Chapter 5 Network and Transport Layers
Top-Down Network Design Chapter Four Characterizing Network Traffic
Network Core and QoS.
Pong: Diagnosing Spatio-Temporal Internet Congestion Properties
Event Building With Smart NICs
Chapter 4 Network Layer Computer Networking: A Top Down Approach 5th edition. Jim Kurose, Keith Ross Addison-Wesley, April Network Layer.
Network Core and QoS.
Presentation transcript:

An Efficient Gigabit Ethernet Switch Model for Large-Scale Simulation Dong (Kevin) Jin

2 Overview and Motivation Gigabit Ethernet High bandwidth (1Gb/s) Widely used to build large-scale networks with many applications Packet delay and packet loss is mainly caused by switch

3 Overview and Motivation Switch Model Fast simulation speed Accurate delay and loss Use simulation to study applications on large-scale Gigabit Ethernet Ethernet Switch … Relay Data Aggregator … …......

4 Switch Models Model Simulation Speed Accuracy of delay/loss Detailed Models SlowVery High E.g. OPNET, OMNet++ Model internal details Different models for different types of switches High computational cost

5 Switch Models Model Simulation Speed Accuracy of delay/loss Detailed Models SlowVery High Simple Queuing Model Very FastLow E.g. Ns-2, DETER Simple FIFO queue One model for everything

6 Switch Models Model Simulation Speed Accuracy of delay/loss Detailed Models SlowVery High Simple Queuing Model Very FastLow [Roman2008] [Nohn2004] FastHigh One model Multiple queues Based on data collected from real switches

7 Switch Models Model Simulation Speed Accuracy of delay/loss Detailed Models SlowVery High Simple Queuing Model Very FastLow [Roman2008] [Nohn2004] FastHigh Our ModelFastestHigh Black-box Switch Model Not model internal details Explore relations between data-in and data-out Based on data from real switches

8 Model Design Approach 1. Perform Experiments on real switch 2. Build Analytical model 3. Build RINSE model 4. Evaluate Simulation Speed and Accuracy

9 Model Design Approach 1. Perform Experiments on real switch 2. Build Analytical model 3. Build RINSE model 4. Evaluate Simulation Speed and Accuracy

10 Experiment Data sequence to collect One-way delay in switch Packet loss pattern in switch Challenges in Gigabit Environment High bit rate - 1Gb/s Small delay in switch -  s

11 Experiment Difficulty Accurate timestamp for one-way delay (  s resolution) Software Timestamp At NIC driver Large delay generated at end hosts at high bit rate (>500Mb/s) Hardware timestamp (NetFPGA) 10 ns resolution Eliminating end-host delay

12 NetFPGA Card 1234 Experiment Setup Constant Bit Rate UDP flows Time_2 - Time_4 = delay per packet inside switch Problem: capture only about 2000 packets without a miss at 1Gb/s Input pcap Time 2Time 4 switch

13 Preliminary Experimental Results - Packet Delay (Low Load) Delay NOT dependent on sending rate Sufficient processing power to handle single flow up to 1Gb/s Model packet delay as a constant Delay Vs Sending Rate (packet size = 100 Bytes)

14 Preliminary Experimental Results - Packet Delay (High Load) Flow1: Mean Delay Vs Sending Rate (packet size = 100 Bytes) Performance varies about 100 times One model not enough for all switches Have to build model based on experimental data flow 1 Background traffic {

15 Preliminary Experimental Results - Packet Loss 0 - received 1 - lost A Packet Loss Sample Pattern 3COM Loss rate NetGear 0.4% 3COM 0.6% Strong autocorrelation exists among neighboring packets

16 Model Design Approach 1. Perform Experiments on real switch 2. Build Analytical model 3. Build RINSE model 4. Evaluate Simulation Speed and Accuracy

17 Packet Loss Model Goal Fast simulation speed Accurate average loss rate Accurate autocorrelation Existing Model - Kth Order Markov Chain Next state depends on previous K packets 2^K states e.g. K=2

received 1 - lost state 1 state 2 state 3 Packet Loss Model Our Markov Chain Model state type 1 - long burst of 0s state type 2 - short burst of 0s state type 3 - burst of 1s Next state depends on Current state #successive packets already in the current state

19 Delay Model - Copula Delay model needs to capture Marginal distribution of packet delay Autocorrelation among neighboring packets Copula Model Sklar’s theorem: Joint Distribution is characterized by Marginal distributions of each component Copula

20 Gaussian Copula - Result Gaussian Copula Given , we can generate the output series {Z t } e.g. for n=3, (Z 1, Z 2, Z 3 )  Z 4, then (Z 2, Z 3, Z 4 )  Z 5,… Same autocorrelation specific by  Each Z t has the marginal distribution ~ N(0,1)  - Standard Gaussian pdf N~(0,1),  - correlation coefficient matrix

21 Model autocorrelation very well Computational efficiency conditional distribution for multivariate Gaussian where Why Gaussian Copula?

22 Gaussian Copula - Apply to Empirical Distribution If the empirical delay distribution X(t) is not Gaussian? (1)Transform X(t) standard Gaussian Y(t), through CDF F(X) = G(Y) (2)Gaussian Copula: Y(t)  Z(t) (3)Transform Z(t) back to W(t), similar to (1) Delay (  s) CDF X W

23 Gaussian Copula - Apply to Empirical Distribution If the empirical delay distribution X(t) is not Gaussian? (1)Transform X(t) standard Gaussian Y(t), through CDF F(X) = G(Y) (2)Gaussian Copula: Y(t)  Z(t) (3)Transform Z(t) back to W(t), similar to (1) Autocorrelation Table L ag

24 Summary Need an efficient switch model to study applications on large-scale gigabit Ethernet A black-box model with focus on Fast simulation speed Accurate delay and loss Experimental results justified our approach Experiment  analytical model  Simulation model  Evaluation

25 Ongoing Work Experiment Collect long data traces with Endace DAG cards Test with cross-interface traffic Model Model for more complicated traffic (TCP, mixing traffic) Correlation between delay and loss Evaluation Compare simulation speed with the other models Compare simulation-generated trace with real data traces

26 Thank You

27 Gaussian Copula Details Autocorrelation,  matrix Conditional distribution for multivariate Gaussian

28 RINSE - Switch Model 3 Models in Switch Layer Our black-box model Simple output queue model Flip-coin model Expected Simulation Time on a chain of switches (about routers): complex queuing model > simple output queuing model > our black-box model ≥ coin model Switch Ethernet MAC Ethernet PHY Switch IP Ethernet MAC Ethernet PHY Host A UDP APP IP Ethernet MAC Ethernet PHY Host B UDP APP

29 Expanding Our Packet Loss Model 1,13,1 1,23,2 Lost 2,1 3,3 3,N 2,2 Received (long burst) Received (short burst) 1,32, ,M ,K......

30 Experimental Results - Packet Delay (High Load) Packet Delay at Beginning of experiment under differenet sending rate (Mb/s) 3COM - Processor Sharing No idea about bit rate until sufficient packets passed Assign max weight at beginning Passed packets   bit rate dertermined  weight   delay 

31 Experimental Results - Packet Delay (Low Load) Single flow Delay NOT depends on sending rate Sufficient processing power to handle 1Gb/s single flow Model packet delay as a constant

32 Experiment Setup with Software Timestamp Host NIC 1NIC 2Sender PortReceiver Port Timestamp process Packet capture process Switch Send to self Timestamp at NIC driver NIC to NIC overhead

33 RINSE - Architecture Large scale network simulation Incorporates hosts, routers, links, interfaces, protocols, etc Domain Modeling Language (DML) A range of implemented network protocols Emulation support DML Configuration SSFNet configure SSF [Simulation Kernel] enhance SSF Standard/API implements Protocol Graph Interface 1 MAC PHY Interface N MAC PHY IPV4 ICMP Emulation Socket TCPUDPDNP3 MODBUS BGPOSPF …

34 Black-Box Testing RFCs 2544 and Guidelines Describing the steps to determine the capabilities of a router No discussing on how to create model from measurements [Hohn 2004] Bridging router performance and queuing theory Simple queuing model, no loss events, and no interactions among ports 12 DAG cards synchronized by GPS [Roman 2007] A black-box router profiler Software testbed (ns2, Click modular router) Focus on single UDP flow and multiple TCP flows

35 [Hohn 2004] “Bridging router performance and queuing theory” Simple output FIFO queue Adding delay based on empirical data before entering the queue No packet loss No interactions among ports

36 [Roman 2008] [Roman 2008] A Device-Independent Router Model Multiple queue model Device-specific parameters derived from experimental data Focus on accurate queue size, number of servers Input ports Output ports

37 Kth order Markov Chain [Yajnik 1999] 2^k states, 2^(k+1) conditional probability Special case: bernoulli model (k=0), two-state markov Chain (k=1) Two-state Markov Chain (Gilbert Model) Extended Gilbert Model Existing Packet Loss Model

38 Autocorrelation of Packet Loss Pattern

39 Latency expectations on wired Ethernet End-to-end latency in switched Ethernet a function of the scheduling and call admission procedure in use not specific to Ethernet hard guarantee the Guaranteed Service specs (RFC 2212) statistical guarantees earliest-deadline-first virtual clock service curve based schedulers (INFOCOM 2000 paper by Liebeherr)