1
play

1 Objective Key Ideas Propose an analysis method for large-scale - PDF document

Oct 24, 2022 •252 likes •294 views

On Scalable Modeling of TCP Congestion Contents Control Mechanism for Large-Scale IP Networks Background Objectives and Key Ideas Fluid-Based Modeling Hiroyuki Ohsaki TCP congestion control mechanism RED router Graduate

  1. On Scalable Modeling of TCP Congestion Contents Control Mechanism for Large-Scale IP Networks � Background � Objectives and Key Ideas � Fluid-Based Modeling Hiroyuki Ohsaki – TCP congestion control mechanism – RED router Graduate School of – Link propagation delay Information Science and Technology � Steady State Analysis Osaka University, Japan � Numerical Examples � Conclusion and Future Works 1 2 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 Background: Conventional Techniques for Background: Large-Scale Networks Networks Analysis � Emergence of large-scale networks � Mathematical analysis – Communication networks is becoming larger and – Queuing theory is a powerful tool for small-scale more complex networks, but... – e.g., network with 10,000 nodes and 100,000 flows – Not applicable to large-scale networks � Urgent need for analysis technique of large-scale � Simulation networks – Several network simulators are available, but... – Ensure stability, reliability, and robustness – Not applicable to large-scale networks – Allow future network expandability and design � Still limited to small-scale networks – Asses impact of network failures and natural disasters 3 4 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 Challenges Possible Solutions � How statistical/dynamical behavior of large-scale � Mathematical analysis networks can be analyzed? – Advanced queuing systems – Statistical behavior � e.g., BCMP network � e.g., throughput, average delay, packet loss – Fluid-based modeling probability � Simulation – Dynamical behavior – Parallel/distributed simulator � e.g., convergence time, ramp-up time, overshoot � e.g., PDNS (Parallel/Distributed NS) � Must be scalable and accurate for complex large-scale – Fluid-based simulation networks � e.g., SSF (Scalable Simulation Framework) 5 6 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 1

  2. Objective Key Ideas � Propose an analysis method for large-scale networks � Extend existing fluid-based modeling approach – Analyze both statistical and dynamical behaviors � Model network components as SISO systems – Applicable to complex closed-loop networks – TCP congestion control mechanism SISO: Single – Scalable to large-scale networks – RED router Input and Single Output – Accurate in diverse network parameters – Link propagation delay � Join SISO systems for building entire network model � Perform steady-state analysis and numerical simulation 7 8 SAINT 2005 SAINT 2005 2005/2/4 2005/2/4 Modeling TCP Congestion Control Block Diagram of TCP Congestion Control Mechanism Mechanism � x(t): input (arrival rate of ACK packets) � y(t): output (transmission rate of data packets) � R: round-trip time � z(t) = y(t - R) - x(t) multiplicative decrease additive increase TCP timeout 9 10 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 Modeling RED Router Block Diagram of RED Router � x(t): input (packet arrival rate) � y(t): output (packet departure rate) � minth, maxth, maxp, wq: RED control parameters packet marking probability 11 12 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 2

  3. Modeling Link Propagation Delay Modeling Entire Network � x(t): input (packet arrival rate) � Connect independent SISO systems � y(t): output (packet departure rate) � Example: case of a single TCP flow and RED router � τ : propagation delay of the link constant delay component 13 14 SAINT 2005 SAINT 2005 2005/2/4 2005/2/4 Steady State Analysis: Main Results Numerical Examples: Network Model TCP packet output rate TCP throughput � � � RED packet output rate � TCP round-trip time � TCP packet loss probability 15 16 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 Numerical Examples: Packet Loss Numerical Examples: Comparison in Probability Diverse Network Parameters our analytic our analytic model shows model shows highest accuracy highest accuracy 300+ sets of parameter configurations 17 18 SAINT 2005 2005/2/4 SAINT 2005 2005/2/4 3

  4. Conclusion Future Works � Proposed an analysis method for large-scale networks � Build analytic model for other network components – Modeled network components as SISO systems – Transport-layer protocols � TCP congestion control mechanism � e.g., TCP Vegas, TCP SACK, HighSpeed TCP, DCCP, TFRC, SCTP � RED router – AQM (Active Queue Management) mechanisms � Link propagation delay � e.g., PI-controller, SRED, DRED, BLUE, FRED – Connect SISO systems for modeling entire network � Implement a fluid-based simulator � Analyzed both statistical and dynamical behaviors – Can simulate large-scale networks – Scalable to large-scale networks – Compatible with ns2 simulation script – Accurate in diverse range of network parameters 19 20 SAINT 2005 SAINT 2005 2005/2/4 2005/2/4 4

Recommend

More recommend