Congestion Control in Satellite Networks
Doctoral Dissertation, Date: October 2007, Advisor: John S. Baras
Due to exponential increases in internet traffic, Active Queue Management (AQM) has been heavily studied by numerous researchers. However, little is known about AQM in satellite networks. A microscopic examination of queueing behavior in satellite networks is conducted to identify problems with applying existing AQM methods. A new AQM method is proposed to overcome the problems and it is validated using a realistic emulation environment and a mathematical model. Three problems that were discovered during the research are discussed in this dissertation.
The first problem is oscillatory queueing, which is caused by high buffering due to Performance Enhancing Proxy (PEP) in satellite networks where congestion control after the PEP buffering does not effectively control traffic senders. Existing AQMs that can solve this problem have tail drop queueing that results in consecutive packet drops (global synchronization). A new AQM method called Adaptive Virtual Queue Random Early Detection (AVQRED) is proposed to solve this problem.
The second problem is unfair bandwidth sharing caused by inaccurate measurements of per-flow bandwidth usage. AVQRED is enhanced to accurately measure per-flow bandwidth usage to solve this problem without adding much complexity to the algorithm.
The third problem is queueing instability caused by buffer flow control where TCP receive windows are adjusted to flow control traffic senders instead of dropping received packets during congestion. Although buffer flow control is quite attractive to satellite networks, queueing becomes unstable because accepting packets instead of dropping them aggravates the congestion level. Furthermore, buffer flow control has abrupt reductions in the TCP receive window size due to high PEP buffering causing more instability. AVQRED with packet drop is proposed to solve this problem.
Networks with scarce bandwidth and high propagation delays can not afford to have an unstable AQM. In this research, three problems that are caused by existing AQMs are identified and a new AQM is proposed to solve the problems. This research can be used by the satellite industry to improve gateway performances and provide better end-user experiences.