Last updated 10/26/2001

Reading List

Notes about the reading list


Internet architecture and philosophy

Specification of the initial TCP protocol. Notice that TCP and IP had not been separated at the time of writing. How did they separate TCP into IP and TCP? Also, the address was 24 bits instead of the current 32 bits. For the specification of current IP and TCP, see [RFC 791] for current IP, and [RFC 793] for current TCP. 

Discuss the ISO/OSI seven layers.

Protocol specification of IPv4.

Protocol specification of TCP.

End to end arguments is a class of system design principles that organize and guide the placement of function within a system. These arguments and the underlying principles have now been invoked in many contexts, becoming part of the vocabulary of network protocol and operating system designers. See [RSC98] and [Cla00] for new comments. See [Ise97] for a very interesting discussion on where to put "intelligence" for networks.

Discuss the design goals of the Internet.

Consider the design and implementation issues of layered network architecture. Propose Application Level Framing (ALF) and Integrated Layer Processing (ILP).

Discuss IPv6.

A summary of the basic principles of Internet. Very easy to read.

An interesting discussion on "stupid" and "intelligent" networks. A strong recommendation to read.

Comments active network research, using the end to end arguments.

"I would argue that we need more than ever to understand [the end-to-end argument] and to apply it as we evolve the network... Future potential is hard to visualize, but our inability to make out the details should not justify locking the doors against new ways to use the network."

Rethinking the end to end arguments. Consider the new requirements for network support.

MAC and link layer

Specification of the initial Ethernet protocol. Notice that it is different from the current protocol.

A summary of the theoretical analyses on Ethernet performance. Measure the performance also.

Improve the congestion control and message exchange of MACA, which was proposed by Kahn and used RTS/CTS to address the hidden terminal problem.

A broad summary of wireless network topics.


Show the pathologies,  stability, and symmetry of Internet routing. Pay attention to the methodology. You may find some other talks by this author also very interesting.

BGP router send too many BGP packets. Why?

Consider the effects of routing policy on convergence.

"The starting point of this work is to step back from the current state, examine how the Internet might develop in the future and derive a new set of requirements for a routing architecture from this work."

Simulate the CPU and bandwidth overhead of OSPF.

Router--high performance

One of the first papers to discuss how to do fast and scalable forwarding table lookup when we use prefix match.

Present MGR, which has five innovations: 1) each forwarding engine has a complete set of the forwarding table (used to have only cache); 2) Use a switched backplane (used to use a shared bus); 3) split forwarding engines from line cards; 4) line cards translate local link layer headers to and from an abstract link layer header format so that forward engines do not need to consider link layer; 5) include QoS; the forwarding engines classify packets, scheduling on the output line cards, in a specialized processor called the QoS processor.

Router--more functionality--active networks

Internet measurements and models (delay, loss, traffic, routing behaviors, topologies etc)

A simple and elegant analysis on packet delay, bandwidth, packet size, and loss.

Present that traffic is self-similar in a local area network environment.

Present that traffic is self-similar in wide area network environments.

Show the pathologies, stability, and symmetry of Internet routing. Pay attention to the methodology. You may find some other talks by this author also very interesting.

Explain LAN and WAN traffic through structral models and show the relationship between self-similarity at the macroscopic level of WANs with the heavy-tailed behavior at the microscopic level of individual sources. 

"Modeling and simulating the Internet is not an easy task. The goal of this paper is to discuss some of the issues and difficulties in modeling Internet traffic, topologies, and protocols."

Traffic management--congestion control, router mechanisms and pricing

Specifies a rate-based, receiver-driven congestion control protocol.

Specifies both router policies and user policies for congestion avoidance (not congestion control). Routers calculate average queue length at regeneration cycle. Users use AIMD. 

The classical paper on TCP congestion control.

Propose the Additive Increase and Multiplicative Decrease algorithm.

First scheduling paper in computer networks.

Compare (by arguments) window control vs. rate control, open-loop control vs. feedback control, router-based vs. source-based, backpressure, reservation vs. walk-in, as well as a single scheme or multiple schemes.

Revision to [Jac88]. Add fast retransmit and fast recovery.

The first paper on binary feedback congestion control.

Congestion control using control theory. The objective is to control the number of outstanding packets at the bottleneck. Use packet pair to estimate the service rate at the bottleneck. Assume round-robin like service at the bottleneck. A different perspective.

Propose and evaluate RED.

The signaling protocol for IntServ.

An overview of the architecture of IntServ. See [DiffServ] for comparison.

One of the proposals for ATM .congestion control.

Propose service curves. Decouple bandwidth and delay.

An interesting modification to TCP/Reno, as proposed in [Jac88, Jac90].

Simulation comparisons of TCP/Tahoe, TCP/Reno, TCP/SACK. Read the new paper in SIGCOMM 2001 to see how they identify TCP versions.

Modify network-layer software at the base station and mobile host, and preserve the end-to-end semantics of TCP.

"The more academic literature has largely focused on devising optimal pricing policies; achieving optimal welfare requires charging marginal congestion costs for usage. In this paper we critique this optimality paradigm ... we contend that the research agenda on pricing in computer networks should shift away from the optimality paradigm and focus more on structural and architectural issues." Here is a list  maintained by Richard Gibbens' on Pricing Internet Resources (Last modified in 1998.).

A summary of TCP congestion control. Refer to the book by the same author.

An overview of the architecture of DiffServ. See [IntServ] for comparison.

How to trick TCP?

This paper discusses issues in the evaluation of TCP performance.

An model of TCP/Reno using stochastic equation.

An model of TCP/Reno using renewal theory.

A refined version of equation-based TCP-friendly congestion control. Notice how they measure loss rate.

Propose GAIMD. Show that if the increase and decrease parameters satisfy a simple relationship, a GAIMD flow is TCP-friendly.

"This document highlights the outstanding architectural issues relating to the deployment and use of QoS mechanisms within internet networks"

Using classic control theory to analyze the performance RED, and show the effects of the parameters.

Study the aggressiveness, responsiveness, smoothness, and fairness of 4 TCP-friendly congestion control protocols by using step functions.

Extension to [YKL01].

How to control high-bandwidth flows to enforce congestion control?

An active queue management scheme based on virtual queue. Compare with RED [FJ93]

Traffic management--adaptive applications (streaming media, games, etc.)

Adaptation by predication.

This document does not directly discuss the ways that middleboxes both enable and hinder the evolution of the Internet architecture, but it lays down the terminology as a first step. 

Network security issues

Measure the cost of SSL.

The first proposal to solve the DDoS problem by IP traceback.

A lower bound on multicast group rekeying.

Traceback using bloom filter.

Consider the effectiveness of router filtering, assuming the network topology satisfies  power-law.

Reliable group rekey.

Beyond unicast: multicast and anycast

The first model on IP multicast.

Propose CBT.

Discuss IP anycast.

Use a random key to control feedback traffic.

Propose PIM.

Propose SRM and use ALF. Notice how they solve feedback implosion.

"For each topic... we describe the state of the art and we present the areas of current active reserach."

Emulate TCP behavior in layered multicast.

An attempt to simplify IP multicast model and address the issues of IP multicast.

"We examine the issues that have limited the commercial deployment of IP-multicast from the viewpoint of carriers."

Partition into clusters. Run TCP among clusters.

Experimental evaluation of end host multicast.

Peer-to-peer networks

Domain Name System

"The original function and purpose of the DNS is reviewed, and contrasted with some of the functions into which it is being forced today and some of the newer demands being placed upon it or suggested for it... This document [is] ... a strong suggestion that the time has come to begin thinking more broadly about the problems we are encountering and possible approaches to solving them."

"The current collective wisdom is that DNSSEC is 1) important, 2) a buzzword, 3) hard, 4) immature. To capture the true state of the technology and identify where work is needed, an informal gathering of groups known to be involved in DNSSEC was held in conjunction with the 49th IETF."


Books (in reverse chronological order)