Computer Networks: A Systems Approach – Free Online Book (Peterson and Davie, 6th Edition) BSCS BSIT

BSCS and BSIT students can read “Computer Networks: A Systems Approach” by Larry Peterson and Bruce Davie free on the authors’ own site. The online edition is the 6th edition (version 6.2-dev), released under a Creative Commons licence, and it covers the standard Computer Networks syllabus: links and Ethernet, switching and IP, routing and BGP, TCP and UDP, congestion control, security and applications.

The systems approach is the reason to pick it. Instead of walking down the OSI layers one by one, each chapter opens with a problem to solve, such as connecting a host to a network or getting two processes to talk, and then builds the mechanisms step by step. Congestion control gets a chapter of its own, and the 6th edition adds QUIC, BBR, software defined networks, gRPC and cloud topics. It works as a semester textbook and as revision before a networking paper, but the online edition has no end-of-chapter exercises.

Book Overview

CourseComputer Networks / Data Communication and Networking
Degree ProgramsBSCS, BSIT
LevelUniversity / Undergraduate (intermediate)
Edition6th Edition, online version 6.2-dev (continuously updated); a 7th edition is in development
AuthorsLarry Peterson and Bruce Davie
CopyrightElsevier, 2012 (original 5th edition content), as stated in the authors’ attribution notice
Online PublisherSystems Approach project (systemsapproach.org)
Print EditionMorgan Kaufmann (Elsevier), 6th edition, sold separately
LanguageEnglish
Total Chapters9 chapters, each ending with a Perspective section; no appendices
ExercisesThe online edition leaves out the print book’s exercise sets; the instructor solutions manual is held by Elsevier
LicenseCreative Commons Attribution 4.0 International (CC BY 4.0)
FormatFree online book (HTML); no official PDF; print and e-book copies are sold separately

Chapter List

Chapter 1: Foundation

Foundational · Difficulty: Easy · Key topics: stakeholders, packet switching, statistical multiplexing, layering, encapsulation, OSI and Internet architecture, sockets, bandwidth, latency, delay × bandwidth product

The chapter starts with a question: how would you build a network that could grow to global size? It looks at everyday applications and then at what three groups need from a network: application developers, network operators and network designers. The requirements are scalable connectivity, cost-effective sharing through packet switching, common services such as reliable channels, and manageability. It then introduces layering, protocols, encapsulation and the OSI and Internet architectures, shows a TCP client and server written with the socket API, and defines bandwidth, latency and the delay × bandwidth product.

Key Points:

  • A network is defined recursively: two or more nodes joined by a physical link, or two or more networks joined by a node.
  • Statistical multiplexing shares links at the granularity of a packet, and each switch stores and forwards packets.
  • The book lists three classes of failure: bit errors (including burst errors), packet-level loss, and node or link failures.
  • Every protocol offers a service interface to the layer above and a peer interface to its counterpart; encapsulation adds a header (and sometimes a trailer) at each layer.
  • The Internet architecture is an hourglass: IP is the narrow waist, with many applications above it and many link technologies below it.
  • Latency = Propagation + Transmit + Queue, and signals travel at about 200,000 km per second in optical fibre.
  • A channel with 50 ms one-way latency and 45 Mbps holds 2.25 million bits, about 280 KB, in flight.

Memory Tip: Think of a hollow pipe. Latency is its length, bandwidth is its width, and delay × bandwidth is its volume.

Common Mistake: Assuming a faster link always means a faster response. For a one-byte request over a 100 ms round trip, 1.5 Mbps and 10 Mbps feel the same, because latency dominates.

Important Questions:

  • What is the delay × bandwidth product and why does it matter? It is the latency multiplied by the bandwidth, the number of bits that can be in flight in the pipe. A sender must keep that many bits outstanding to keep the link busy.
  • Why do computer networks use packet switching with statistical multiplexing? Links are shared in small units, so a flow uses capacity only when it has data to send. This is cheaper and more flexible than reserving a circuit for each conversation.

Chapter 2: Direct Links

Core · Difficulty: Medium · Key topics: encoding, framing, CRC, checksum, stop-and-wait, sliding window, Ethernet, CSMA/CD, Wi-Fi, passive optical network, cellular · Dated: SONET, PPP and HDLC are classic examples; Wi-Fi coverage stops at 802.11ac

Chapter 2 asks how one node can be connected to another, or to a network. A network adaptor turns bits into signals using NRZ, NRZI, Manchester and 4B/5B encodings, and framing marks where each frame starts and ends in PPP, HDLC and SONET. Error detection uses the Internet checksum and the CRC, and reliable delivery uses stop-and-wait, sliding window and concurrent logical channels. The second half covers shared media: Ethernet, Wi-Fi and Bluetooth, and the fibre and cellular access networks that ISPs use to reach customers.

Key Points:

  • NRZ sends each bit as a signal level, but long runs of identical bits make clock recovery hard; Manchester sends the XOR of the data and the clock, and 4B/5B inserts extra bits at 80 percent efficiency.
  • HDLC marks frames with 01111110, and the sender inserts a 0 after every five consecutive 1s (bit stuffing).
  • A SONET frame is 810 bytes (9 × 90) and an STS-1 link runs at 51.84 Mbps.
  • The Internet checksum adds 16-bit words in ones’ complement arithmetic; a CRC treats the message as a polynomial, and CRC-32 is widely used.
  • With sliding window, the sender tracks SWS, LAR and LFS; when the receive window equals the send window, SWS < (MaxSeqNum + 1)/2.
  • An Ethernet frame holds up to 1500 bytes of data with 48-bit addresses; after a collision the adaptor doubles its waiting time (exponential backoff).
  • 802.11 uses CSMA/CA with an optional RTS-CTS exchange to cope with hidden nodes; a PON is a tree that can reach up to 1024 homes.

Memory Tip: Framing finds the edges of a frame, the CRC finds the damage, and sliding window repairs the loss.

Common Mistake: Treating bit rate and baud rate as the same thing. With Manchester encoding the bit rate is half the baud rate, because each bit needs two signal changes.

Important Questions:

  • Why does HDLC use bit stuffing? The end-of-frame pattern 01111110 could also appear inside the data. The sender inserts a 0 after five 1s and the receiver removes it, so the marker is never confused with data.
  • What three roles can the sliding window algorithm play? It delivers frames reliably over an unreliable link, it keeps frames in order, and it provides flow control so the sender cannot overrun the receiver.

Chapter 3: Internetworking

Core · Difficulty: Medium · Key topics: datagram, virtual circuit, learning bridge, spanning tree, VLAN, IP, subnetting, CIDR, ARP, DHCP, ICMP, RIP, OSPF, SDN · Dated: ATM (the book calls it well past its peak)

This chapter shows how to join links into networks that reach beyond a single Ethernet. It begins with switching basics (datagrams, virtual circuits with ATM as the example, and source routing) and then switched Ethernet with learning bridges, the spanning tree algorithm and VLANs. The longest section is IP: the service model, header, fragmentation, global addresses, subnetting and CIDR, ARP, DHCP, ICMP and tunnels. Routing follows with distance-vector (RIP) and link-state (OSPF) algorithms and link metrics, and the chapter ends with software and hardware switch design and software defined networks.

Key Points:

  • A datagram carries the full destination address and needs no setup; a virtual circuit sets up state in every switch and uses a VCI looked up in a VC table.
  • A learning bridge records the source address and arrival port of each frame; the spanning tree algorithm elects the switch with the smallest ID as root and switches off redundant ports to stop loops.
  • IP offers best-effort delivery: each network has an MTU, so large datagrams are fragmented and reassembled only at the destination.
  • Subnet masks and CIDR prefixes such as 192.4.16/24 let routers aggregate routes, and forwarding uses the longest matching prefix.
  • ARP maps IP addresses to link-layer addresses, DHCP configures hosts, and ICMP reports errors and sends redirects.
  • Distance-vector routing (Bellman-Ford, RIP) shares vectors with neighbours and can count to infinity; link-state routing floods link-state packets and runs Dijkstra’s algorithm (OSPF).
  • SDN separates the control plane from the data plane and runs forwarding on bare-metal switches.

Memory Tip: Forwarding is a table look-up done for every packet; routing is the slower work that builds the table.

Common Mistake: Mixing up Ethernet and IP addresses. Ethernet addresses are flat and unique, while IP addresses are hierarchical and carry a network part, which is why ARP is needed to map one to the other.

Important Questions:

  • What is the difference between forwarding and routing? Forwarding means receiving a packet, looking up its destination in a table and sending it on. Routing is the process, run by routing protocols, that builds those tables.
  • How does a link-state protocol learn the network? Each node floods a link-state packet describing its links and costs to every other node. Each node then has the full map and runs Dijkstra’s algorithm to compute the best route to every destination.

Chapter 4: Advanced Internetworking

Advanced · Difficulty: Hard · Key topics: autonomous system, OSPF areas, BGP, IPv6, multicast, PIM, MPLS, Mobile IP · Dated: DVMRP and the MBone (history)

Chapter 4 is about scale. It describes the global Internet as a set of autonomous systems and shows how OSPF areas and BGP add hierarchy to routing. BGP is explained as a path-vector protocol where policy, not shortest distance, decides routes. IPv6 follows with 128-bit addresses, a simpler header, extension headers, autoconfiguration and anycast. Multicast covers group addresses, DVMRP, PIM sparse mode and source-specific multicast. MPLS adds labels, explicit routing and VPNs, and the last section solves routing to mobile hosts with Mobile IP.

Key Points:

  • An autonomous system is a single administrative domain; OSPF-style protocols run inside it and BGP runs between autonomous systems.
  • BGP advertisements carry the complete AS path, so a router that sees its own AS in a path can reject it and avoid a loop.
  • Common AS relationships are provider-customer, customer-provider and peer, and each has its own advertisement policy.
  • An OSPF routing domain can be split into areas, and the backbone area (area 0) connects them.
  • IPv6 addresses are 128 bits (about 3.4 × 10^38), written as eight 16-bit hexadecimal blocks, with runs of zeros compressed.
  • DVMRP builds a broadcast tree and prunes it; PIM sparse mode lets receivers send Join messages towards a rendezvous point.
  • MPLS forwards by short fixed-length labels, and Mobile IP uses a home address, a home agent and a care-of address.

Memory Tip: DVMRP prunes down from a flooded tree, while PIM-SM grows a tree up through explicit joins.

Common Mistake: Believing BGP picks the shortest path. BGP only tries to find some policy-compliant route, and policy often follows who pays whom.

Important Questions:

  • Why does BGP carry the full AS path? It lets a router detect loops, because it will see its own AS in the path, and it gives policies the information they need to choose between routes.
  • What is triangle routing in Mobile IP and how is it reduced? Packets from a correspondent go to the home network first and are then tunnelled to the mobile host. Route optimization lets the correspondent learn the care-of address and tunnel directly.

Chapter 5: End-to-End Protocols

Core · Difficulty: Hard · Key topics: UDP, TCP, three-way handshake, flow control, Nagle, adaptive retransmission, QUIC, RPC, gRPC, RTP · Dated: SunRPC and DCE-RPC are design examples; old Xeon test machines

Chapter 5 turns host-to-host delivery into process-to-process communication. UDP is the simple demultiplexor with ports and an optional checksum. TCP takes most of the chapter: segment format, the three-way handshake and state diagram, sliding window with flow control, wraparound, triggering transmission with Nagle’s algorithm, and adaptive retransmission through the Karn/Partridge and Jacobson/Karels algorithms. It also covers record boundaries, SCTP and QUIC. RPC is explained with SunRPC, DCE-RPC and gRPC, and RTP and RTCP serve real-time media.

Key Points:

  • UDP identifies the target process by port number, and its checksum covers the data and a pseudoheader.
  • The three-way handshake agrees on starting sequence numbers; a connection moves through states such as CLOSED, ESTABLISHED, FIN_WAIT and TIME_WAIT.
  • TCP flow control uses a 16-bit AdvertisedWindow, and the 32-bit sequence space wraps in 6.4 hours at 1.5 Mbps but in about 3 seconds at 10 Gbps.
  • Nagle’s algorithm uses the arrival of an ACK like a timer, sending small segments only when earlier data has been acknowledged.
  • EstimatedRTT = alpha × EstimatedRTT + (1 - alpha) × SampleRTT; Karn/Partridge ignores samples from retransmitted segments, and Jacobson/Karels adds a deviation term to the timeout.
  • RPC can offer at-most-once semantics using message IDs, and gRPC runs over HTTP/2.
  • The first 12 bytes of the RTP header are always present, and the sequence number and timestamp help the receiver play media smoothly.

Memory Tip: The handshake is SYN, SYN+ACK, ACK: each side proves it has heard the other side’s starting sequence number.

Common Mistake: Thinking TCP keeps message boundaries. It is a byte stream, so 8, 2 and 20 bytes written by the sender may be read as six reads of 5 bytes.

Important Questions:

  • Why does TCP need an explicit connection set-up phase? A TCP connection joins two processes on any two hosts, and old packets may still be in the network. The handshake lets both sides agree on initial sequence numbers and allocate state.
  • What problem did the Karn/Partridge algorithm fix? An ACK after a retransmission is ambiguous, so the RTT sample could be wrong. TCP now takes RTT samples only from segments sent once, and doubles the timeout after each retransmission.

Chapter 6: Congestion Control

Advanced · Difficulty: Hard · Key topics: resource allocation, fair queuing, AIMD, slow start, fast retransmit, CUBIC, RED, ECN, BBR, DCTCP, IntServ, DiffServ · Dated: IntServ and RSVP were not widely deployed

Chapter 6 asks how a network shares its bandwidth and buffers fairly and effectively. It first builds a taxonomy: router-centric or host-centric, reservation-based or feedback-based, and window-based or rate-based, judged by throughput, delay and fairness. FIFO with tail drop and fair queuing come next. TCP congestion control is explained through AIMD, slow start, fast retransmit and fast recovery, and CUBIC. Advanced congestion control covers DECbit, RED, ECN, Vegas, BBR and DCTCP, and the chapter closes with quality of service using IntServ with RSVP and DiffServ.

Key Points:

  • Congestion collapse hit the Internet in the 1980s when hosts sent as fast as the advertised window allowed; Van Jacobson added congestion control to TCP.
  • MaxWindow = MIN(CongestionWindow, AdvertisedWindow), so a sender never exceeds what the network or the receiver can take.
  • AIMD adds about one packet per RTT and halves the congestion window on loss.
  • Slow start doubles the window each RTT from one packet until the congestion threshold is reached.
  • Fast retransmit resends a segment after three duplicate ACKs instead of waiting for a timeout.
  • RED drops packets randomly once the average queue length passes MinThreshold; ECN marks packets instead of dropping them.
  • IntServ reserves resources per flow with RSVP and soft state, while DiffServ marks packets into a few classes with per-hop behaviours such as EF and AF.

Memory Tip: Slow start is not slow. The name only means it starts from one packet, and then it doubles every RTT.

Common Mistake: Confusing flow control with congestion control. Flow control protects the receiver through the advertised window; congestion control protects the network through the congestion window.

Important Questions:

  • How does TCP detect and react to congestion? It treats a timeout or three duplicate ACKs as a sign of loss caused by congestion, halves the congestion window, and then probes for more capacity by increasing the window again.
  • What is the difference between IntServ and DiffServ? IntServ reserves resources for individual flows using RSVP, which has scalability problems. DiffServ gives treatment to a small number of traffic classes marked in the packet header.

Chapter 7: End-to-End Data

Advanced · Difficulty: Medium · Key topics: presentation formatting, marshalling, XDR, ASN.1, Protocol Buffers, XML, compression, JPEG, MPEG, MP3, adaptive streaming · Dated: XDR, ASN.1 and NDR are older formats

Chapter 7 is about what the bytes in a message mean. Presentation formatting explains how data types are marshalled for the network, comparing a canonical intermediate form with receiver-makes-right, and tagged with untagged encodings, using XDR, ASN.1, NDR and Protocol Buffers, then XML with schemas and namespaces. Multimedia covers lossless compression (run length, DPCM and Lempel-Ziv), JPEG with its DCT, quantization and encoding phases, MPEG video with I, P and B frames, adaptive streaming over HTTP, and MP3 audio.

Key Points:

  • Hosts differ in byte order (big-endian or little-endian), so the sender and receiver must agree on one data format.
  • XDR uses big-endian 32-bit integers without tags, ASN.1 BER encodes each item as tag, length and value, and Protocol Buffers use numbered fields from a .proto file.
  • Run length encoding turns AAABBCDDDD into 3A2B1C4D; DPCM stores differences, and Lempel-Ziv builds a dictionary (the idea behind gzip).
  • JPEG applies an 8 × 8 DCT, then lossy quantization, then lossless zigzag run length and delta encoding.
  • MPEG uses I frames (self-contained), P frames (predicted from earlier frames) and B frames (predicted from both directions), with typical compression around 90:1.
  • HTTP adaptive streaming lets a client choose between several encodings of a video using a manifest.
  • CD-quality audio is sampled at 44.1 kHz with 16 bits per sample, and MP3 compresses it using a model of human hearing.

Memory Tip: In MPEG, I is Independent, P is Predicted from the past, and B looks in Both directions.

Common Mistake: Calling the whole JPEG pipeline lossy. The DCT and the encoding phase are lossless; only quantization throws information away.

Important Questions:

  • What are the two conversion strategies for marshalling data? Canonical intermediate form converts everything to one agreed external format, as XDR does. Receiver-makes-right sends data in the sender’s format with a tag, and the receiver converts it, as NDR does.
  • Why does MPEG compress better than JPEG on every frame? It also exploits similarity between frames, using P and B frames with motion estimation, so most frames store only the changes from their neighbours.

Chapter 8: Network Security

Advanced · Difficulty: Hard · Key topics: threats and trust, ciphers, public keys, authenticators, certificates, Diffie-Hellman, Kerberos, TLS, IPsec, firewalls · Dated: TLS section still lists MD5, SHA-1 and DES-era algorithms

Chapter 8 begins with the point that security means managing trust and threats, and that perfect security does not exist. The cryptographic building blocks are secret-key and public-key ciphers and authenticators such as hashes, HMACs and digital signatures. Key predistribution covers certificates, certification authorities, the web of trust, revocation and Diffie-Hellman. Authentication protocols defend against replay with nonces and timestamps, with Needham-Schroeder and Kerberos as examples. The example systems are PGP, SSH, TLS, IPsec, Wi-Fi security and firewalls.

Key Points:

  • The security goals are confidentiality, integrity, authentication, access control and availability.
  • The cipher functions are public and only the key is secret, a principle first stated in 1883.
  • Secret-key ciphers such as AES are fast but need N(N-1)/2 keys for N parties; public-key ciphers such as RSA give each party one key pair.
  • An authenticator (cryptographic hash, HMAC or digital signature) checks both integrity and authenticity, which encryption alone does not.
  • A certificate signed by a certification authority binds an identity to a public key; PGP uses a web of trust, and a CRL revokes bad certificates.
  • Diffie-Hellman creates a shared secret over an open channel but does not authenticate the participants.
  • TLS has a handshake protocol to negotiate ciphers and keys and a record protocol to protect data; firewalls cannot stop an attacker who is already inside.

Memory Tip: With public keys, you lock with the public key and unlock with the private key; for a signature, you sign with the private key and anyone verifies with the public key.

Common Mistake: Believing that encryption alone proves who sent a message or that it was not changed. Without an authenticator, a tampered ciphertext can still decrypt into valid-looking text.

Important Questions:

  • What is a replay attack and how can it be stopped? An adversary resends a copy of an old valid message. Including a timestamp or a nonce under the authenticator lets the receiver reject messages that are not fresh.
  • What does a certification authority do? It verifies an identity and signs a certificate that binds that identity to a public key. Anyone who knows the CA’s own key can then trust the binding.

Chapter 9: Applications

Core · Difficulty: Medium · Key topics: SMTP, MIME, IMAP, HTTP, REST, SOAP, SIP, DNS, SNMP, overlay networks, BitTorrent, CDN · Dated: SOAP, WSDL and H.323 are history and contrast

The last chapter argues that applications need their own protocols. Traditional applications cover email with SMTP, MIME and IMAP, and the web with HTTP, URIs, persistent connections, caching, HTTP/3 over QUIC, and the shift from SOAP and WSDL to REST and cloud services. Multimedia applications use SDP, SIP and H.323. Infrastructure applications are DNS name resolution and network management with SNMP and OpenConfig. The final section covers overlay networks: routing overlays, end system multicast, resilient overlays, peer-to-peer networks such as BitTorrent, and content distribution networks.

Key Points:

  • An email message has a header and a body in ASCII (RFC 822), MIME carries other data types, SMTP moves mail between hosts, and IMAP retrieves it.
  • An HTTP request has a start line with method, URL and version; response codes fall into five classes by their first digit.
  • HTTP 1.1 added persistent connections, and HTTP/3 runs over QUIC.
  • SOAP and WSDL define a custom protocol for each service; REST uses the small set of HTTP methods such as GET and POST and puts the complexity in the payload.
  • DNS is a hierarchical name space read right to left, divided into zones, each served by two or more name servers, with caching to speed up resolution.
  • SNMP provides GET and SET on MIB variables, while OpenConfig supports automated, zero-touch management.
  • Structured peer-to-peer overlays hash object names to nodes; BitTorrent splits a file into pieces shared in a swarm; a CDN uses redirectors to send each request to a nearby server.

Memory Tip: Read a DNS name from right to left: for penguins.cs.princeton.edu, resolution starts at edu, then princeton, then cs.

Common Mistake: Mixing up an application program and an application protocol. A browser is a program; HTTP is the protocol it speaks, and many different programs can speak it.

Important Questions:

  • How does DNS resolve a name such as penguins.cs.princeton.edu? A client asks a name server starting at a root server, which points to the server for the next zone, until the authoritative server returns the address. Answers are cached to avoid repeating the walk.
  • How do SOAP/WSDL and REST differ? SOAP/WSDL define a different protocol for each application. REST reuses the standard HTTP methods for every service and moves the complexity into the data being exchanged.

How to Study This Book

The book has nine chapters and no appendices, and it is meant to be read in order. The Preface groups them for you: Chapter 1 is foundational, Chapters 2, 3, 5 and 9 are core, and Chapters 4, 6, 7 and 8 are advanced material that may or may not fit into a single semester. Inside each chapter the sections also move from basic ideas to specific technologies to advanced techniques, so you can stop at a section boundary without losing the thread.

For a first networking course, read Chapter 1 completely and do the arithmetic yourself: latency, the delay × bandwidth product and the 50 ms by 45 Mbps example. Then take Chapter 2 up to Section 2.6 (encoding, framing, CRC, sliding window and Ethernet) and treat the Wi-Fi and access network sections 2.7 and 2.8 as reading for context. Chapter 3 is the heart of the course: spend your time on spanning tree, the IP header, fragmentation, subnetting and CIDR, and the two routing algorithms in Section 3.4. Work the distance-vector and Dijkstra examples on paper.

Chapter 5 is the other must-study chapter, because TCP is central to almost every networking course. Read UDP in ten minutes, then go slowly through the three-way handshake, the state diagram, the advertised window and the adaptive retransmission formulas. Sections 5.3 (RPC) and 5.4 (RTP) can be skimmed in a first pass. Chapter 6 depends on Chapter 5, so read Section 6.3 on TCP congestion control right after it, and come back to queuing disciplines and quality of service later. Chapter 9 is the easiest to read and can be taken earlier if you want a feel for HTTP, DNS and email before the harder middle chapters.

Chapters 4, 7 and 8 are best treated as electives unless your course outline lists them. In Chapter 4, BGP in Section 4.1 and IPv6 in Section 4.2 matter most; multicast, MPLS and Mobile IP are specialised. Chapter 7 mixes data formats and media compression, so read only the section your syllabus names. Chapter 8 is a good short introduction to cryptography and secure protocols, but it assumes you are comfortable with Chapter 5, and the TLS section is easier to follow after you have used HTTPS in Chapter 9.

Some parts have aged and are worth knowing about. The book itself calls ATM “well past its peak”, SONET, PPP, HDLC, DVMRP, SunRPC, DCE-RPC, SOAP and H.323 are shown mainly as design examples, the Wi-Fi coverage stops at 802.11ac with 802.11ax as emerging, and the authors say the code samples are no longer compiled or checked, so treat them as a guide. None of this affects the core chapters on IP, TCP, congestion control and DNS.

The online edition has no end-of-chapter exercise sets, and the instructor solutions manual for the print book is controlled by the publisher. Make your own practice: compute subnet masks, trace RIP updates, draw a TCP timeline with a lost segment, and write the client and server from Chapter 1 in C or Python. The book describes ping and traceroute but has no Wireshark packet-capture labs, so run Wireshark, ping, traceroute and dig beside the chapters to see the protocols you read about.


Used In These Programs

This book is used for the Computer Networks course in: BS Computer Science · BS Information Technology. Browse all Computer Networks books or all Computer Science books.

Who Should Read This

This book is written for BSCS and BSIT students taking a first or second Computer Networks course, usually taken after a first programming course and a data structures course. It assumes you can read simple C-style code and are comfortable with basic data structures and operating system ideas, but it needs no earlier networking knowledge. It is also a good choice for anyone who wants to understand why the Internet works as it does, not just memorise protocol headers. It is not the right book if you need hands-on router configuration, Cisco-style certification practice or a first introduction to computers. For a second view with exercises and a protocol-by-protocol approach, read Computer Networking: Principles, Protocols and Practice next to it, and for the operating system side of sockets and buffers, Operating Systems: Three Easy Pieces and Think OS are useful companions.


Applicable Universities

This book is useful for students at Pakistani universities offering BS Computer Science or BS Information Technology with a computer networks course, including Punjab University, Virtual University, COMSATS, FAST, UET, NUST, and other HEC-recognized institutions.

Official Copy of Computer Networks: A Systems Approach

This book is free from its official site. The authors publish it as an online edition and do not offer an official PDF, so the legitimate free way to read it is the complete online book on book.systemsapproach.org, chapter by chapter. Click the button below to open the official site. Print and e-book copies are sold separately.

→ Visit Official Page

FAQs

Is Computer Networks: A Systems Approach free?

Yes. The complete online edition is free to read at book.systemsapproach.org, and the source is released under the Creative Commons Attribution 4.0 International licence (CC BY 4.0), which allows copying and reuse with credit. There is no official PDF on the site, so use the online edition rather than a file reposted elsewhere. The printed textbook from Morgan Kaufmann (Elsevier) is a separate paid product.

Which edition is this?

The online book is labelled 6th Edition, and its current version is 6.2-dev, a continuously updated release. Version 6.0 (August 2019) roughly equals the 6th edition, version 6.1 was tagged in November 2019, and the Preface is dated October 2020. The authors say they are now writing a 7th edition in a separate project, so the 6th edition remains the full free book for now.

How many chapters and exercises does it have?

There are 9 chapters: Foundation, Direct Links, Internetworking, Advanced Internetworking, End-to-End Protocols, Congestion Control, End-to-End Data, Network Security and Applications. Each ends with a Perspective section. The online edition does not include the exercise sets of the print book, and the instructor solutions manual is held by the publisher, so you will need to make your own practice problems.

Is it good for complete beginners?

It is aimed at students who have finished a first programming course and know basic data structures, because the code samples are in C and the later chapters build on earlier ones. You do not need earlier networking knowledge. If Chapter 1 feels heavy, read it twice and work the bandwidth and latency examples by hand before moving to Chapter 2.

Does it cover TCP, congestion control, BGP, IPv6 and TLS?

Yes. TCP and UDP are in Chapter 5, with the handshake, sliding window, Nagle and adaptive retransmission, and congestion control is in Chapter 6 with AIMD, slow start, CUBIC, RED, ECN and BBR. BGP, IPv6, multicast and MPLS are in Chapter 4, and TLS, IPsec and firewalls are in Chapter 8. DNS, HTTP and email are in Chapter 9.

Is it still current in 2026?

Mostly. The core ideas behind IP, TCP, congestion control, DNS and the web have not changed, and the 6th edition adds QUIC, HTTP/3, BBR, SDN, gRPC and cloud networking. Some parts show their age: the Wi-Fi section stops at 802.11ac with 802.11ax as emerging, ATM, SONET, DVMRP, SOAP and H.323 are covered as history, and the authors note that the code samples are no longer compiled or checked. Use the book for the ideas and add current vendor or RFC documentation where your course needs it.

Related Books

Computer Networks: A Systems Approach teaches the Internet by asking what problem each mechanism solves, from Ethernet frames to TCP congestion control and TLS, all in one free online book. Browse more Computer Science books for the rest of your semester.

Computer Networks: A Systems Approach (6th Edition, online version 6.2-dev) by Larry Peterson and Bruce Davie. Licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). Copyright: Elsevier, 2012, as stated in the authors’ attribution notice. Source: https://github.com/SystemsApproach/book. Official online book: https://book.systemsapproach.org/. This page links to the official site and does not host or copy the book.

Leave a Comment