A First-Principles Approach to Networked Systems

Author

Affiliation

UC Santa Barbara — Systems and Networking Lab (SNL)

Published

March 31, 2026

Preface

Why First Principles

For decades, advanced networking has been taught as a protocol survey: a chapter per topic, a page per protocol, students leaving the course able to describe TCP and CSMA/CA but unable to reason about a protocol they have not yet seen. The cost of that approach was always borne quietly: when QUIC arrived, students had to learn it from scratch. Generative AI has now made that cost untenable. An LLM produces a correct-sounding TCP walkthrough in seconds. Assignments and exams that test protocol production are AI-trivializable. What survives is the skill the protocol survey was never designed to teach: recognizing when a confident-sounding claim about an unfamiliar system is structurally wrong.

This book organizes networking around four invariants every networked system must answer, regardless of era or technology, and teaches students to use those invariants to evaluate claims and derive architectures from constraints.

The Framework

The framework has four components.

Four invariants define what every system must answer:

State: What does the system know about its environment, and where does that knowledge live?
Time: What cadence drives feedback, event-triggered or timer-triggered, prescribed or inferred?
Coordination: Who decides? One entity, many, or a hierarchy?
Interface: What is exposed and what is hidden?

The questions are invariant; the answers differ. ALOHA, TCP, BGP, 802.11ax’s centralized scheduler, and modern agentic operations all become legible as different answers to the same four questions, forced by different constraints in their environments.

Environment–Measurement–Belief decomposition turns why systems fail into a three-layer diagnostic that transfers across distance-vector routing, ABR streaming, and bufferbloat: failures are gaps between what is true (Environment), what the system observes (Measurement), and what it concludes (Belief).

The binding-constraint cascade is the predictive tool. A binding constraint can lock more than one invariant answer at once, which forces the rest into cascade. A student who has internalized the cascade can derive what an unseen architecture must look like from its constraints, and reject claims that violate the derivation.

Disaggregation and closed-loop reasoning are the two moves for constructing an answer once a binding constraint narrows the space. Disaggregation separates concerns so each can evolve independently, the choice behind IP’s minimal interface and behind SDN’s split between control plane and data plane. Closed-loop reasoning asks whether a chosen set of answers, taken together, converges or oscillates, the question behind TCP’s ACK-clocked feedback loop and behind bufferbloat’s failure to signal congestion in time.

Chapter 1 works through that check on TCP, DNS, DHCP, and WiFi; the chapters that follow apply it to every system the book covers.

How the Book Is Organized

The book follows the Internet’s layered architecture bottom-up. Each chapter is a pioneer arc: successive protocols emerge as answers to specific failures of the previous generation under specific binding constraints, tracked through invariant analysis, E-M-B decomposition, and the binding-constraint cascade. ALOHA’s collision problem motivates carrier sensing; the contention ceiling motivates centralized scheduling; spectrum economics force WiFi and cellular onto opposite ends of the coordination spectrum, then converge again at 802.11ax.

Part I: Foundations

Chapter 1 (First Principles): introduces the four invariants, the E-M-B decomposition, and the binding-constraint cascade through worked examples (TCP, DNS, DHCP, WiFi) that recur throughout the book.

Part III: The Network Layer (Forwarding, Routing, and Queue Management)

The IP datagram is the narrow waist. Everything below delivers it; everything above uses it.

Chapter 5 (Addressing, Naming, and Discovery): covers IP addressing, DNS, classless interdomain routing, DHCP and ARP, network address translation, and the transition to IPv6.
Chapter 6 (Routing, Switching, and Programmable Networks): moves from distance-vector and link-state routing through BGP, MPLS, software-defined networking, and P4-programmable data planes.
Chapter 7 (Queue Management): traces queue management from FIFO and fair queueing through RED, the bufferbloat problem, and its resolution in CoDel and FQ-CoDel.

Part IV: The Transport Layer (End-to-End Delivery)

The IP interface forces transport to solve reliability and congestion control from endpoints only.

Chapter 8 (Transport and Congestion Control): follows transport and congestion control from Cerf and Kahn’s original design through Jacobson’s congestion-avoidance fix, CUBIC, BBR, DCTCP, and QUIC.
Chapter 9 (System Composition): examines how transport and queue management interact, covering L4S, PowerBoost, and the distinction between QoS and QoE.

Part V: The Application Layer (Services on the Stack)

Applications inherit the transport interface and solve their own invariant problems.

Chapter 10 (Application Protocols and Content Delivery): traces HTTP’s evolution alongside content delivery networks, QUIC and HTTP/3, and edge computing.
Chapter 11 (Multimedia Applications): covers real-time multimedia from RTP through adaptive streaming (DASH and HLS), model-predictive control, WebRTC, and low-latency streaming.

Part VI: Cross-Cutting Concerns

Chapter 12 (Measurement, Management, and the Research Frontier): surveys network measurement and management from BPF-based instrumentation through streaming telemetry, in-band network telemetry, Sonata, and network verification.

Part VII: Agentic Systems (Capstone)

Chapter 13 (Agentic Systems): closes the book by applying the four invariants to agentic AI systems, tracing the progression from prompting through chain-of-thought reasoning, the ReAct pattern, multi-agent coordination, the Model Context Protocol as a thin waist, and persistent memory. The capstone shows the framework applying to the newest system class: agents built on the infrastructure the preceding chapters describe.

How to Use This Book

Each chapter opens with the binding constraint that shapes the system under study. It walks through pioneer-driven Acts (chronological, each with invariant analysis and E-M-B tables) and closes with a Grand Arc revealing cross-generational patterns.

The book serves upper-division CS undergraduates and beginning graduate students who have completed an introductory networking course. It assumes familiarity with TCP/IP, basic queuing, and protocol layering, not how to reason about these systems.

Work in Progress

This book is under active development alongside CS 176C at UC Santa Barbara. Chapter drafts are not currently public; please contact the author for review access.

Arpit Gupta, UC Santa Barbara Spring 2026

--- title: "Preface" number-sections: false --- # Preface {.unnumbered} ## Why First Principles For decades, advanced networking has been taught as a protocol survey: a chapter per topic, a page per protocol, students leaving the course able to describe TCP and CSMA/CA but unable to reason about a protocol they have not yet seen. The cost of that approach was always borne quietly: when QUIC arrived, students had to learn it from scratch. Generative AI has now made that cost untenable. An LLM produces a correct-sounding TCP walkthrough in seconds. Assignments and exams that test protocol *production* are AI-trivializable. What survives is the skill the protocol survey was never designed to teach: recognizing when a confident-sounding claim about an unfamiliar system is structurally wrong. This book organizes networking around four invariants every networked system must answer, regardless of era or technology, and teaches students to use those invariants to evaluate claims and derive architectures from constraints. ## The Framework The framework has four components. **Four invariants** define what every system must answer: - **State**: What does the system know about its environment, and where does that knowledge live? - **Time**: What cadence drives feedback, event-triggered or timer-triggered, prescribed or inferred? - **Coordination**: Who decides? One entity, many, or a hierarchy? - **Interface**: What is exposed and what is hidden? The questions are invariant; the answers differ. ALOHA, TCP, BGP, 802.11ax's centralized scheduler, and modern agentic operations all become legible as different answers to the same four questions, forced by different constraints in their environments. **Environment–Measurement–Belief decomposition** turns *why systems fail* into a three-layer diagnostic that transfers across distance-vector routing, ABR streaming, and bufferbloat: failures are gaps between what is true (Environment), what the system observes (Measurement), and what it concludes (Belief). **The binding-constraint cascade** is the predictive tool. A binding constraint can lock more than one invariant answer at once, which forces the rest into cascade. A student who has internalized the cascade can derive what an unseen architecture must look like from its constraints, and reject claims that violate the derivation. **Disaggregation and closed-loop reasoning** are the two moves for constructing an answer once a binding constraint narrows the space. Disaggregation separates concerns so each can evolve independently, the choice behind IP's minimal interface and behind SDN's split between control plane and data plane. Closed-loop reasoning asks whether a chosen set of answers, taken together, converges or oscillates, the question behind TCP's ACK-clocked feedback loop and behind bufferbloat's failure to signal congestion in time. Chapter 1 works through that check on TCP, DNS, DHCP, and WiFi; the chapters that follow apply it to every system the book covers. ## How the Book Is Organized The book follows the Internet's layered architecture bottom-up. Each chapter is a **pioneer arc**: successive protocols emerge as answers to specific failures of the previous generation under specific binding constraints, tracked through invariant analysis, E-M-B decomposition, and the binding-constraint cascade. ALOHA's collision problem motivates carrier sensing; the contention ceiling motivates centralized scheduling; spectrum economics force WiFi and cellular onto opposite ends of the coordination spectrum, then converge again at 802.11ax. ### Part I: Foundations **Chapter 1** (*First Principles*): introduces the four invariants, the E-M-B decomposition, and the binding-constraint cascade through worked examples (TCP, DNS, DHCP, WiFi) that recur throughout the book. ### Part II: The Link Layer (Sharing a Physical Medium) The binding constraint inherited from the physical layer: electromagnetic spectrum is finite, shared, and noisy. - **Chapter 2** (*Wired Link Layer*): traces Ethernet from CSMA/CD contention through the spanning tree protocol, switched forwarding, fat-tree datacenter fabrics, and VXLAN/EVPN overlays. - **Chapter 3** (*Wireless Link Layer*): follows wireless medium access from ALOHA and CSMA/CA through 802.11ax's OFDMA scheduling, cellular CDMA, and massive MIMO. - **Chapter 4** (*Wireless Infrastructure*): traces cellular infrastructure from the monolithic base transceiver station through LTE's eNodeB, 5G's disaggregated CU/DU/RU architecture, O-RAN, and network slicing. ### Part III: The Network Layer (Forwarding, Routing, and Queue Management) The IP datagram is the narrow waist. Everything below delivers it; everything above uses it. - **Chapter 5** (*Addressing, Naming, and Discovery*): covers IP addressing, DNS, classless interdomain routing, DHCP and ARP, network address translation, and the transition to IPv6. - **Chapter 6** (*Routing, Switching, and Programmable Networks*): moves from distance-vector and link-state routing through BGP, MPLS, software-defined networking, and P4-programmable data planes. - **Chapter 7** (*Queue Management*): traces queue management from FIFO and fair queueing through RED, the bufferbloat problem, and its resolution in CoDel and FQ-CoDel. ### Part IV: The Transport Layer (End-to-End Delivery) The IP interface forces transport to solve reliability and congestion control from endpoints only. - **Chapter 8** (*Transport and Congestion Control*): follows transport and congestion control from Cerf and Kahn's original design through Jacobson's congestion-avoidance fix, CUBIC, BBR, DCTCP, and QUIC. - **Chapter 9** (*System Composition*): examines how transport and queue management interact, covering L4S, PowerBoost, and the distinction between QoS and QoE. ### Part V: The Application Layer (Services on the Stack) Applications inherit the transport interface and solve their own invariant problems. - **Chapter 10** (*Application Protocols and Content Delivery*): traces HTTP's evolution alongside content delivery networks, QUIC and HTTP/3, and edge computing. - **Chapter 11** (*Multimedia Applications*): covers real-time multimedia from RTP through adaptive streaming (DASH and HLS), model-predictive control, WebRTC, and low-latency streaming. ### Part VI: Cross-Cutting Concerns - **Chapter 12** (*Measurement, Management, and the Research Frontier*): surveys network measurement and management from BPF-based instrumentation through streaming telemetry, in-band network telemetry, Sonata, and network verification. ### Part VII: Agentic Systems (Capstone) - **Chapter 13** (*Agentic Systems*): closes the book by applying the four invariants to agentic AI systems, tracing the progression from prompting through chain-of-thought reasoning, the ReAct pattern, multi-agent coordination, the Model Context Protocol as a thin waist, and persistent memory. The capstone shows the framework applying to the newest system class: agents built on the infrastructure the preceding chapters describe. ## How to Use This Book Each chapter opens with the **binding constraint** that shapes the system under study. It walks through **pioneer-driven Acts** (chronological, each with invariant analysis and E-M-B tables) and closes with a **Grand Arc** revealing cross-generational patterns. The book serves upper-division CS undergraduates and beginning graduate students who have completed an introductory networking course. It assumes familiarity with TCP/IP, basic queuing, and protocol layering, not how to *reason* about these systems. ::: {.callout-warning} ## Work in Progress This book is under active development alongside [CS 176C](https://sites.cs.ucsb.edu/~arpitgupta/cs176c/spring26/) at UC Santa Barbara. Chapter drafts are not currently public; please contact the author for review access. ::: --- *[Arpit Gupta](https://sites.cs.ucsb.edu/~arpitgupta/), UC Santa Barbara* *Spring 2026* --- *© 2026 Arpit Gupta, UC Santa Barbara. All rights reserved. Contact the author for permissions.*