To: Peter Ecclesine who wrote (21147 ) 4/28/2007 9:19:20 AM From: Peter Ecclesine Respond to of 46821 Hi All, From a Doug Comer talk this January on the Internet Economics and QoS Lesson 12: Quality-of-Service (QoS) offers no substantial benefit over best-effort service if underlying pipes have sufficient capacity, and cannot improve service for everyone if pipes have insufficient capacity. Achieving Reliable Transfer Lesson 14: Only one layer of a protocol stack should handle reliability. Lesson 15: Reliability should be provided end-to-end, not link-by-link. The Future of Layer 2 Protocols Lesson 16: Although it is difficult to predict details, we know that no matter how it works, the next Local Area Network technology will be called Ethernet. Protocols and Heterogeneity Lesson 17: A protocol design that accommodates all possible underlying networks without bias allows engineers to experiment with new packet switching technologies and permits competition to flourish. [e.g. sensor networks, surveillance networks] Aggressive Retransmission Lesson 18: Although intuition suggests that aggressive retransmission will increase throughput, many aggressive senders cause congestion collapse. Addressing Model Lesson 19: The options of using an address to identify an endpoint or using an address to identify a network connection each have advantages and disadvantages. Identifying an endpoint makes mobility possible, but requires per-host routing and identification; identifying a network connection provides security, but makes mobility difficult. End-to-end Addressing Lesson 20: End-to-end addressing is not as critical as we once thought; over one-half of all computers on the Internet are now located behind a NAT box. Permutations of layers Lesson 22: For any protocols x and y, it is possible to transfer x over y, regardless of layering. Interesting examples: -- PPPoE (Point-to-Point Protocol over Ethernet), -- VPNs using IP over IP, -- iSCSI (bus protocol over IP) Understanding The Internet Lesson 30: Building a large packet-switching network is easy; understanding the nature of traffic in a large packet-switching network is impossible. Optimizing The Internet Lesson 31: No optimal Internet architecture exists; the best we can expect is a multivariate pareto optimization. Managing The Internet Lesson 32: Anyone who has a large network, has a large network management problem. Routing In The Internet Lesson 33: Routing is among the most intellectually challenging aspects of interworking; after thirty years of research, networks often exhibit problems such as route flapping and black holes. The Triumph Of Packet Switching Lesson 34: A viable, efficient, global communication network can be created using a connectionless packet-based technology that follows a best-effort paradigm.