CongestionWindowControlAlgorithms幻灯片.pptVIP

Congestion Window Control Algorithms David Wei For Pizza Talk @ Netlab Topics History Component of Congestion Window Control Algorithm FAST (sc2002 version) Synchronization Model Layered Architecture based on Time Scale Open Issues History of TCP Congestion Control John Nagle’s Lixia Zhang’s, Karn’s Algorithm DEC-bit, Tahoe, CARD Tri-S, DUAL Reno Vegas New Reno HSTCP, Scalable TCP, FAST, Westwood … ??? Raj Jain’s “Three Questions that a congestion control algorithm should answer”: How frequent to make decision Decision function How to Increase/Decrease Window How Frequent? Time Scale of a Congestion control algorithm Slow Start Per Packet (acknowledgement) Congestion Avoidance Per RTT Rate Halving Per Loss … … Layered-Architecture based on Time Scale Functionality Layer Smooth change of CWND Per packet Decision based on Feed back Per RTT Decision based on Feed back; Severe congestion avoidance Per Loss Decision Function Decision Function under Binary Signal Congestion Signal Delay Loss ECN Loss-Based Delay-Based Binary Signal Reno DUAL Multi-Valued Signal TFRC Vegas Congestion Signal Transition Point Non-Transition Point Increase/Decrease Function Under Synchronization Model (Chui and Jain): Oscillation Increase/Decrease Function Binary Signal Yes AIMD Multi-Valued Signal Flexible Equation-based Congestion Control Algorithms Algorithm Signal I/D Function Tahoe (New) Reno Loss; Binary AIMD HSTCP Loss; Binary Aggressive AIMD Scalable TCP Loss; Binary MIMD CARD Delay; Binary AIMD DUAL Delay; Multi-Valued AIMD Vegas Delay; Multi-Valued Loss; Binary Equation Based AIMD+AIAD FAST (Ver. sc2002) Delay; Multi-Valued Loss; Binary Equation Based AIMD+Exponential FAST (sc2002) Algorithm Do the following every other RTT: per acknowledgement: Calculate: If w(t)w*: w(t)=w(t)-1 If w(t)w*: w(t)=w(t)+1 Or Randomize the Increase/Decrease Synchronization Model Assume: All the sources make the decision slow enough that the queues converge whenever the sources make its