Register your product to gain access to bonus material or receive a coupon.
Groundbreaking Patterns for Building Simpler, More Powerful Networks
In Patterns in Network Architecture, pioneer John Day takes a unique approach to solving the problem of network architecture. Piercing the fog of history, he bridges the gap between our experience from the original ARPANET and today’s Internet to a new perspective on networking. Along the way, he shows how socioeconomic forces derailed progress and led to the current crisis.
Beginning with the seven fundamental, and still unanswered, questions identified during the ARPANET’s development, Patterns in Network Architecture returns to bedrock and traces our experience both good and bad. Along the way, he uncovers overlooked patterns in protocols that simplify design and implementation and resolves the classic conflict between connection and connectionless while retaining the best of both. He finds deep new insights into the core challenges of naming and addressing, along with results from upper-layer architecture. All of this in Day’s deft hands comes together in a tour de force of elegance and simplicity with the annoying turn of events that the answer has been staring us in the face: Operating systems tell us even more about networking than we thought. The result is, in essence, the first “unified theory of networking,” and leads to a simpler, more powerful–and above all–more scalable network infrastructure. The book then lays the groundwork for how to exploit the result in the design, development, and management as we move beyond the limitations of the Internet.
Using this new model, Day shows how many complex mechanisms in the Internet today (multihoming, mobility, and multicast) are, with this collapse in complexity, now simply a consequence of the structure. The problems of router table growth of such concern today disappear. The inescapable conclusion is that the Internet is an unfinished demo, more in the tradition of DOS than Unix, that has been living on Moore’s Law and 30 years of band-aids. It is long past time to get networking back on track.
• Patterns in network protocols that synthesize “contradictory” approaches and simplify design and implementation
• “Deriving” that networking is interprocess communication (IPC) yielding
• A distributed IPC model that repeats with different scope and range of operation
• Making network addresses topological makes routing purely a local matter
• That in fact, private addresses are the norm–not the exception–with the consequence that the global public addresses required today are unnecessary
• That mobility is dynamic multihoming and unicast is a subset of multicast, but multicast devolves into unicast and facilitates mobility
• That the Internet today is more like DOS, but what we need should be more like Unix
• For networking researchers, architects, designers, engineers
Provocative, elegant, and profound, Patterns in Network Architecture transforms the way you envision, architect, and implement networks.
Preface: The Seven Unanswered Questions xiii
Chapter 1: Foundations for Network Architecture 1
Chapter 2: Protocol Elements 23
Chapter 3: Patterns in Protocols 57
Chapter 4: Stalking the Upper-Layer Architecture 97
Chapter 5: Naming and Addressing 141
Chapter 6: Divining Layers 185
Chapter 7: The Network IPC Model 235
Chapter 8: Making Addresses Topological 283
Chapter 9: Multihoming, Multicast, and Mobility 317
Chapter 10: Backing Out of a Blind Alley 351
Appendix A: Outline for Gedanken Experiment on Separating Mechanism and Policy 385
Bibliography 389
Index 399
Listen to a Podcast CS Techcast 27: Taking You to Network School with John Day
Patterns in Network Architecture: Naming and Addressing
Preface: The Seven Unanswered Questions xiii
Chapter 1: Foundations for Network Architecture 1
Chapter 2: Protocol Elements 23
Chapter 3: Patterns in Protocols 57
Chapter 4: Stalking the Upper-Layer Architecture 97
Chapter 5: Naming and Addressing 141
Chapter 6: Divining Layers 185
Chapter 7: The Network IPC Model 235
Chapter 8: Making Addresses Topological 283
Chapter 9: Multihoming, Multicast, and Mobility 317
Chapter 10: Backing Out of a Blind Alley 351
Appendix A: Outline for Gedanken Experiment on Separating Mechanism and Policy 385
Bibliography 389
Index 399
Print Number | Error Location | Error | Correction | Date Added |
---|---|---|---|---|
2 | p xx | Splitting out IP (nothing new for addressing). Splitting IP from TCP seemed a necessity. The transport protocol and IP do very different functions (as will become clear in Chapter 6, Diving Layers). The only unfortunate aspect in the creation of IP was that nothing was done about the multihoming problem. IP continued to name the interface. But this was understandable. IP was split out in 1975, soon after the problem was recognized. Although we understood what the multihoming problem was and theoretically what its solution was, there was still much about addressing that was unclear. More theoretical and practical work was necessary. However, it did put us in the uncomfortable position of an Internet address naming a subnetwork point of attachment. |
Splitting out IP (nothing new for addressing). Splitting IP from TCP seemed a necessity. The transport protocol and IP do very different functions (as will become clear in Chapter 6, Divining Layers). The only unfortunate aspect in the creation of IP was that nothing was done about the multihoming problem. IP continued to name the interface. But this was understandable. IP was split out in 1975, soon after the problem was recognized. Although we understood what the multihoming problem was and theoretically what its solution was, there was still much about addressing that was unclear. More theoretical and practical work was necessary. However, it did put us in the uncomfortable position of an Internet address naming a subnetwork point of attachment. |
5/8/2008 |
2 | p xxvi | That brings us to roughly the early 1990s, to the time frame when I started this exercise, just as the IPng was heating up.26 The seven unanswered questions we started with were still unanswered and in the back of my mind (as they always had been). It was not my intention to try to solve them. It is a daunting list. But with each that pattern emerged, was measured against whether they contributed to solving them. I was looking for a clear understanding of where we were. However, three issues had to be looked at. Two of the issues experience had shown could wreck an architecture if not confronted and solved. We have already touched on them: finding a meaningful synthesis of connection and connectionless, and working out naming and addressing (and in particular what location dependent means). The religious war over connections and connectionless had been at the root of too many disasters. A true synthesis was desperately needed. And, of course, just looking at the seven unanswered questions, you can see that a number of issues all revolve around a clear understanding of naming and addressing. The third arose from my experience with hundreds of protocol designs more than 20 years, seeing the same things over and over. I wanted to separate mechanism and policy as we had in operating systemsjust to see what would happen.27 |
That brings us to roughly the early 1990s, to the time frame when I started this exercise, just as the IPng was heating up.26 The seven unanswered questions we started with were still unanswered and in the back of my mind (as they always had been). It was not my intention to try to solve them. It is a daunting list. But as each pattern emerged, it was measured against whether they contributed to solving them. I was looking for a clear understanding of where we were. However, three issues had to be looked at. Two of the issues experience had shown could wreck an architecture if not confronted and solved. We have already touched on them: finding a meaningful synthesis of connection and connectionless, and working out naming and addressing (and in particular what location dependent means). The religious war over connections and connectionless had been at the root of too many disasters. A true synthesis was desperately needed. And, of course, just looking at the seven unanswered questions, you can see that a number of issues all revolve around a clear understanding of naming and addressing. The third arose from my experience with hundreds of protocol designs over more than 30 years, seeing the same things over and over. I wanted to separate mechanism and policy as we had in operating systemsjust to see what would happen.27 |
5/8/2008 |
2 | p xxviii | As noted earlier, several concepts that are key to understanding this model are not generally known. We will rely heavily on what Seymour Papert29 calls the only concepts that make computer science worth learning: problem decomposition, abstraction, and recursion. Abstraction has fallen into to disuse for the past couple of decades, but we will put it to good use here. Furthermore, the architecture we are led to requires a considerable cognitive shift. Therefore, this book is organized to take the reader from what we know to a new way of looking at things. To bridge the gap, so to speak. Even so, this will not be easy for the reader; there is some hard thinking ahead. We first start with a return to fundamentals, to remind us of the minimum assumptions required for communication and for the tools for working with abstractions. In Chapters 2 and 3, we look at the familiar world of protocols and separating mechanism and policy. Here, new patterns emerge that indicate there are probably only three kinds of protocols, and then later we find that one of them is more a common header than a protocol. We are also able to make considerable progress in resolving the conflict between connections and connectionless.30 In Chapter 4, we review our experience with upper layers and learn some things that we did right and some things to avoid. As strange as it might sound, we find some key concepts here that will be useful in constructing our fundamental model, while at the same time concluding that there is no upper-layer architecture. Then in Chapter 5, Naming and Addressing,we take a hard look at that ever-difficult and subtle topic, naming and addressing. We give special emphasis to Saltzers 1982 paper expanding on it slightly, noting how the current infatuation with the loc/id split problem is a dead end. By the time we reach Chapter 6, we have a pretty reasonable picture of the problem and the elements we will need and can consider the problem of assembling them into a system. Here we embark on a simple exercise that any of us could have done at any time over the past 30 years only to find it yields the structure we have been looking for. (A revolting department!) This chapter is key to everything. |
As noted earlier, several concepts that are key to understanding this model are not generally known. We will rely heavily on what Seymour Papert29 calls the only concepts that make computer science worth learning: problem decomposition, abstraction, and recursion. Abstraction has fallen into disuse for the past couple of decades, but we will put it to good use here. Furthermore, the architecture we are led to requires a considerable cognitive shift. Therefore, this book is organized to take the reader from what we know to a new way of looking at things. To bridge the gap, so to speak. Even so, this will not be easy for the reader; there is some hard thinking ahead. We first start with a return to fundamentals, to remind us of the minimum assumptions required for communication and for the tools for working with abstractions. In Chapters 2 and 3, we look at the familiar world of protocols and separating mechanism and policy. Here, new patterns emerge that indicate there are probably only three... |