IAI@Rome Summer 2005
Networking Tutorial (August 3, 2005)
Title: TCP: Making it Fat, Fast, and Unfair
Speaker: Paul Francis
Slides
This tutorial describes the operation of TCP, including Round-Trip Time(RTT) estimation, slow start and Additive Increase Multiplicative Decrease (AIMD) congestion control, fast retransmit and recovery, Selective Acknowledgement (SACK), Explicit Congestion Notification (ECN), and Random Early Detection (RED). We'll discuss common performance problems when TCP is operated in challenging environments, including lossy wireless links and high bandwidth-delay product links such as satellites. Finally, we'll describe several TCP-related research projects at Cornell. These include how to manipulate TCP congestion control algorithms to get differential QoS, how to fully utilize link bandwidth with low-priority "background" applications, and some new ideas on how to speed-up TCP over very fast networks.
Title: Adaptive Group Communication Protocols
Speaker: Robbert van Renesse
Slides
Group communication protocols support fault-tolerant and secure applications. In order to find the right trade-off between performance and robustness, it is important that such protocols can be tailored, both a priori and at run-time, to both networking and application behavior. In order to enable this, it is useful to structure such protocols from small components. Doing so simplifies development, increases dependability, and supports configuration.
The tutorial describes an architecture that supports componentized protocols and formal reasoning, how it may be reconfigured at runtime without disturbing running applications, and how it can provide similar or better performance than monolithic, non-adaptive group communication protocols.
Title: Gossip or Epidemic Protocols
Speaker: Robbert van Renesse
Slides
Gossip protocols, aka epidemic protocols, are randomized protocols that have been used for content dissemination, membership tracking, sensor aggregation, and more. As compared to deterministic protocols, gossip protocols have interesting scaling, robustness, timeliness, and security properties, but come with probabilistic rather than deterministic guarantees and some overhead costs. The tutorial discusses a variety of services based on gossip, and highlights different gossip techniques.