April 10, 2006, Northwestern University eXCP ExCP Offense by Amit April 10, 2006, Northwestern University
Congestion Control Design Space End Point Router Support Try and Backoff TCP, Vegas, RAP, Fast TCP, S-TCP High Speed TCP DecBit, ECN, RED, AQM Request and Set PCP ATM, XCP, WFQ, RCP
Deployment Not end-to-end solution Modification required in routers Anderson et. al. PCP:Efficient Endpoint Congestion control Incremental deployment is infeasible All routers must be XCP enabled What if bottleneck router is not XCP enabled? Loss due to congestion but XCP assumes otherwise
Performance Comparison with TCP-Reno instead of TCP-SACK, TCP-Vegas Only Simulation based evaluation Limitation of sender buffer/receiver window How useful is for short web-traffic? Congestion avoidance vs slow-start
Security Sender Centric vs Receiver Centric Fake H_rtt Fake H_cwnd (H_throughput) Discard H_feedback Compromised/Misbehaving Router Can completely shut-off good flows
Contribution Stoica's CSFQ: Core Stateless Fair Queuing (SIGCOMM '98) Edge routers – per flow management, rate estimation (r) Core routers – fair allocation of bandwidth among flows (f) Each packet of a flow with arrival rate r is forwarded with probability p = min(1, f/r) Core routers estimate fair allocation of b/w using iterative algorithm XCP – modification of CSFQ, worse from the perspective that it cannot detect misbehaving hots nor can do policy enforcement