A Study of VoIP Gateway Selection Techniques Matthew Caesar, Dipak Ghosal, Randy Katz {mccaesar,
Motivation/Goals Goals: Few blocked calls Good Quality of Service (QoS) Preferential QoS to high priority users Methods: Advertising gateway congestion Monitoring and recording call QoS Call redirection Questions: Is it feasible to route based on dynamic congestion information? How best to redirect calls with congestion and QoS sensitivity?
System Architecture Example Call Setup Example Advertisement Gateway User Location Server Internet Admin. Domain
Assumptions ADs cooperate to provide service. Use IETF’s TRIP architecture to support interoperability. Number of Gateways is small enough to store per-ITG routing information at the LS. Cost of multi-gateway-hop routes (e.g. IP- PSTN-IP) will be prohibitively expensive. Majority of QoS degradation is inter-AD, not intra-AD.
Redirection Techniques Local Redirection Random Redirection CS Redirection QoSS Redirection Hybrid Redirection (CSQoSS) cost = alpha*C+(1-alpha)*QoS Price Sensitive Schemes Also varied “setsize” Perform random redirection over the set
Evaluation: Metrics Number of blocked calls Average call QoS Used Mean Opinion Score (MOS) based on RTP loss rate Economic efficiency Ratio of service tier to QoS achieved Stability: Variance in gateway utilization Over time Over the set of gateways Overhead: Advertisement bandwidth consumption
System Performance Results: Call QoS More congestion sensitivity decreases quality Small setsizes offer best quality Results: Blocking Probability More congestion sensitivity decreases blocking probability Small setsizes cause few blocked calls
System Performance Results: Background Traffic QoS sensitivity minimizes effects of cross traffic Even a little sensitivity to QoS vastly improves call quality Results: System Stability More congestion sensitivity decreases blocking probability Small setsizes cause few blocked calls
Effect of Flash Crowd Change from 0.1 to 4.0 offered load in 1 simulated hour Some randomization is necessary to improve the call blocking rate Randomization not required for smaller flash crowds
Conclusions QoS sensitivity necessary for best-effort Internet Suffers from load oscillations Congestion sensitivity necessary for limited gateway resources Chooses poor-QoS paths Hybrid scheme is very effective Good QoS, low blocking probability Resilient to flash-crowds, background traffic Future work Price sensitive schemes Multi-class user model Effective admission control