Skip to content

Latest commit

 

History

History
586 lines (451 loc) · 15.9 KB

2020-09-15-computer-networking-note.md

File metadata and controls

586 lines (451 loc) · 15.9 KB
Raw

Computer Networking

2020-09-15

Computer Networking - A Top Down Approach, 7th

Chapter 1. Computer networks and the Internet

1.2 The Network Edge

  • end systems
    • sit at the edge of the Internet
    • also referred to as hosts, because they host applications

1.2.1 Access Networks

  • access networks
    • the network physically connects an end system to the first router(also known as edge router)

Chapter 2. Application Layer

2.1 Principles of Network Applications

  • network application architectures

    • Client/Server
    • Peer to Peer

  • client and server processes

    • client, the process initiates the communication
    • server, the process waits to be contacted to begin the session

  • socket

    • software interface
    • API between the application and the network
    • developer has little control of the transport-layer side of the socket

  • addressing processes

    • IP for hosts
    • port number for processes

  • transport services available to applications

    • reliable data transfer, also there are unreliable protocols
    • throughput
    • timing
    • security

  • TCP

    • connection-oriented service
    • reliable data transfer service

  • UDP

    • connectionless
    • unreliable

2.2 The Web and HTTP

  • HTTP

    • use TCP as underlying transport protocol
    • stateless protocol
    • HTTP with non-persistent and persistent connections

  • HTTP message format

    • 2 types, request and response

  • HTTP request message format

    • \r\n as separator
    • first line is request line
    • then header lines
    • then blank line
    • then entity body

  • HTTP response message format

    • status line
    • header lines
    • entity body

  • cookie

    • has 4 components
      • cookie header line in HTTP response
      • cookie header line in HTTP request
      • cookie file in end system managed by browser
      • back-end database at the website

  • web caching

    • also called a proxy server
    • CDN, content distribution network
    • conditional GET
      • header If-Modified-Since
      • header Last-Modified
      • HTTP/1.1 304 Not Modified

2.3 Electronic Mail in the Internet

  • protocols used between mail servers

    • SMTP

  • mail access protocols

    • POP3, extremely simple
    • IMAP
    • HTTP, web-based e-mail

2.4 DNS

  • DNS protocol

    • distributed databases
    • application layer protocol
    • runs over UDP, port 53
    • additional functionality
      • host aliasing, canonical hostname
      • mail server aliasing
      • load distribution

  • DNS servers

    • distributed, hierarchical servers
    • 3 classes
      • root DNS servers
      • top-level domain(TLD) servers
      • authoritative DNS servers
    • local DNS server, forward query into the DNS hierarchy
    • query
      • recursive queries
      • iterative queries
    • DNS caching

  • DNS records

    • resource records
    • type
      • A, standard hostname, value is IP
      • NS, domain, value is hostname of authoritative DNS server
      • CNAME, canonical name
      • MX, canonical name of mail server

  • DNS vulnerabilities

    • DDoS, a deluge of DNS query
    • DNS poisoning

2.5 Peer-to-Peer file distribution

(not too much to say)

2.6 Video Streaming and CDN

  • DASH

  • CDN

    • why CDN
      • clients too far from data center
      • same bytes sent again and again to same network
      • single point of failure

(skip)

2.7 Socket programming

  • UDP
  • TCP

Chapter 3. Transport Layer

3.1 Introduction and Transport-Layer Services

  • transport layer

    • provide logical communication
    • implemented in end systems, not network routers
    • extends network layer's delivery services from host-to-host to process-to-process

  • UDP

    • delivery segments
    • simple error checking

  • TCP

    • reliable data transfer
    • congestion control

3.2 Multiplexing and Demultiplexing

  • demultiplexing

    • transport-layer segments with some fields to identify the receiving socket

  • multiplexing

    • gathering data chunks from different sockets then pass the segments to the network layer

  • connectionless multiplexing and demultiplexing

    • UDP
    • process identified by a two-tuple (dest IP, dest port)

  • connection-oriented multiplexing and demultiplexing

    • TCP
    • process identified by a four-tuple (source IP, source port, dest IP, dest port), for simultaneous connections

3.3 Connectionless Transport: UDP

  • why UDP

    • finer application level control over how data sent
      • TCP has congestion control
      • real-time applications
    • no connection establishment, no delay
    • no connection state, less system resources usage
    • small packet header overhead
      • TCP 20 bytes, UDP 8 bytes

  • running multimedia applications over UDP is controversial

    • no congestion control, network would be high stressed, crowding out normal TCP traffics

  • reliable data transfer over UDP

    • reliability done by application itself, google's QUIC

3.4 Principles of Reliable Data Transfer

  • the layer below reliable data transfer protocol may be unreliable

  • rdt 1.0, rdt means reliable data transfer, is a toy protocol

    • use finite-state machine(FSM) definitions
    • data transfer over a perfectly reliable channel

  • rdt 2.0

    • with bit errors
    • ARQ(automatic repeat request) protocols
      • error detection
      • receiver feedback, with positive(ACK)/negative(NAK) packets
      • retransmission
    • stop-and-wait protocols
    • fatal, what if ACK/NAK is corrupted

  • rdt 2.1, handle ACK/NAK corrupted

    • resent packet if checksum of ACK/NAK isn't match
    • add sequence number to packet header to detect retransmission

  • rdt 2.2, send ACK/NAK with sequence number

  • rdt 3.0, over a lossy channel with bit errors

    • how to detect packet loss, and what to do
      • detect on sender side
      • if no reply coming from receiver, after wait long enough, retransmit the packet
      • how long should sender wait? at least round-trip delay
      • may have duplicate data packets, handled by packet sequence number in rdt 2.2
      • countdown timer

  • pipelined transfer protocol

    • the range of sequence number must increase
    • sender and receiver may have more buffer
    • error recovery
      • Go-Back-N
      • selective repeat

  • Go-Back-N protocol

    • window size, N
    • sliding-window protocol
    • sequence number category
      • already ACKed
      • sent, not ACKed
      • usable, not sent yet
      • not usable
    • three types of events
      • invocation from above, when window is full, buffered requests or have a synchronization mechanism
      • receipt of ACK, cumulative acknowledgement
      • a timeout event, resend all packets
    • receiver
      • ACK n-th packet when (n-1)th packet is received
      • discard out of order packets, for simplicity, so receiver don't need buffer out of order packets

  • selective repeat(SR)

    • a single packet error may cause GBN retransmit large number of packets, especially window size is large
    • receiver buffer out of order packets
    • receiver ACK every packet even if it's not in order
    • sender
      • each packet has its timer, resend packet if it is timeout
    • receiver
      • have a receiver window size
      • receive a packet isn't window's lower bound, buffer it, otherwise move forward the window
      • re ACK packets even its ACKed, with certain number below the window's lower bound
    • window size must be less than or equal to half of the maximum sequence number
      • otherwise receiver can't know a packet is a new packet or a retransmission

  • packets arrive out of order

    • sender assume packet can't live longer than a maximum lifetime

3.5 Connection-Oriented Transport: TCP

  • TCP connection

    • a logical connection
    • full duplex
    • three-way handshake

  • TCP send data

    • send buffer
    • MSS, maximum segment size, maximum payload size a TCP segment can hold
    • MTU, maximum transmission unit, largest link-layer frame size

  • TCP receive data

    • receive buffer

  • TCP segment structure

    • (skip)

  • sequence numbers and acknowledgement numbers

    • TCP views data as ordered stream of bytes
    • sequence number is byte number of all bytes to send
    • ACK number is next byte number except to receive

  • estimate RTT (round-trip time)

    • SampleRTT for latest mesurement
    • EstimatedRTT for weighted average RTT
    • DevRTT for difference between SampleRTT and EstimatedRTT
    • SampleRTT fluctuates, EstimatedRTT smooth it
    • DevRTT as margin for fluctuations
    • TCP's timeout is EstimatedRTT + 4 * DevRTT

  • doubling the timeout interval

    • if timeout, set next timeout as twice

  • fast retransmit

    • receiver detects missing segments, gap in data stream, then it ACK for missing segments immediately
    • sender receive same ACKs repeatedly, knowing that segment lost, retransmit it, even before timeout

  • flow control

    • a speed matching service
    • sender maintain a receive window, rwnd, means how much the spare buffer size is on the receiver
    • receiver tell rwnd every time it send segments back to sender
    • sender calculate through ACKed sequence number and sent sequence number
    • if rwnd = 0, sender still keep sending small(1 byte) segments, so receiver can ACK, in which contain new rwnd
    • UDP doesn't have this, if buffer overflow, following segments would be dropped

  • connection management

    • three way handshake
    • TCP state transition diagram
    • SYN cookie, prevent SYN flood attack

3.6 Principles of Congestion Control

  • cause and cost

    • large queuing delays as packet arrival rate near the link capacity
    • large delays cause sender to retransmit, which is unneeded, and make link more congested
    • packet dropping during congestion, which wasted the forward work of upstream routers

  • approaches

    • end-to-end congestion control, without network layer's assistance
    • network-assisted congestion control

3.7 TCP Congestion Control

  • limit sender send rate

    • if little congestion, increase send rate
    • if congestion, decrease send rate

  • cwnd

    • cwnd size is how much data sender send in every RTT

  • congestion control algorithm(TCP reno)

    • slow start
      • add 1 MSS to cwnd after receive per ACK, which means doubling cwnd every RTT, exponential growth
      • if loss event, set ssthresh to cwnd/2, then set cwnd to init value (1 MSS)
      • if cwnd meet ssthresh, enter congestion avoidance mode
    • congestion avoidance
      • cwnd increase 1 MSS per RTT, linear growth
    • fast recovery
      • retransmit when receive 3 duplicated ACK
      • TCP reno use fast recovery

  • retrospective

    • an additive-increase, multiplicative-decrease(AIMD) algorithm
    • alternative congestion control algorithms

  • average throughput

    • is relate to window size and RTT

  • high bandwidth links

    • maximum transfer speed is limited because of congestion control

  • fairness

    • two links with same window size and RTT, will finally come to same transfer rate
    • links with different RTT, smaller RTT will get higher throughput
    • UDP doesn't have congestion control, so UDP traffic will crowd out TCP traffic, unfair
    • parallel TCP connections is unfair, too

  • Explicit Congestion Control(ECN)

    • network layer assisted

Chapter 4. The Network Layer: Data Plane

4.1 Overview of Network Layer

  • network layer functions

    • forwarding, in data plane
      • fast, typically hardware implemented
      • forwarding table
    • routing, in control plane
      • slow, often software implemented

  • SDN

4.2 What's Inside a Router

  • four components

    • input ports
    • switching fabric
    • output ports
    • routing processor

  • forwarding

    • destination-based forwarding
    • generalized forwarding

  • input port processing

    • forwarding by prefix of destination address
    • longest prefix matching rule
    • lookup is hardware implemented, which is super fast

  • switching

    • via memory
    • via a bus
    • via an interconnection network

  • output port processing

(skip)

4.3 IPv4, Addressing, IPv6, and More

  • IPv4 datagram format

    • data length, rarely larger than 1500 bytes because of Ethernet frame size
    • protocol, indicate which protocol used in payload
    • source and destination IP addresses

  • IPv4 datagram fragmentation

    • MTU, maximum transmission transmission, largest link-layer frame size
    • IP datagram may transfer to different link-layer protocol with smaller MTU, where fragmentation happend
    • each IP datagram identified by source address, dest address, identifier in header, offset in header

  • IPv4 addressing

    • IP address is associated with an interface, not a host, a host can have multiple IP address
    • address format, 32 bits long
    • subnet, subnet mask
    • CIDR, classless interdomain routing, identify subnet by prefix
    • ICANN, organization who allocate IP addresses

  • DHCP, dynamic host configuration protocol

    • obtain IP address automatically, also additional infos like, default gateway(first hop router), local DNS, subnet mask
    • four steps
      • DHCP server discovery
      • DHCP server offer
      • DHCP request
      • DHCP ACK

  • NAT

  • IPv6

    • important changes
      • expanded address capabilities
        • 128 bits
        • anycast
      • compact header
      • flow
    • difference with IPv4
      • fragmentation, IPv6 not allow intermediate routers do fragmentation
        • if MTU changed, router drop that datagram and send back "packet too big" ICMP message
        • speed up forwarding
      • header checksum, IPv4 without it
        • don't need recalc after TTL changed, speed up
      • options, so header size is fixed

  • transition from IPv4 to IPv6

    • tunneling, whole IPv6 datagram as payload of IPv4

4.4 Generalized Forwarding and SDN

  • OpenFlow

(skip)

Chapter 5. The Network Layer: Control Plane

5.1 Introduction

  • per-router control
  • logically centrialized control

5.2 Routing Algorithms

  • classify

    • centrialized or decentrialized
    • static or dynamic
    • load-sensitive or load-insensitive

  • link-state algorithm

    • centrialized routing algorithm
    • Dijkstra

  • distance-vector algorithm

    • decentrialized routing algorithm

(skip)

5.3 OSPF

  • AS, autonomouse system

    • a group of routers under same administrative control

  • OSPF, Open Shortest Path First

    • intra-AS routing protocol

5.4 BGP

  • BGP

    • routing among ISPs

  • IP anycast

(skip)

5.6 ICMP, Internet Control Message Protocol

  • function

    • error report
    • ping

  • traceroute

    • sending ICMP messages with TTL 1,2,3,...,etc, to a unreachable port number
    • routers in path send back ICMP warning, destination send back ICMP unreachable

(skip)

(Chapter 6 & 7 read Chinese edition)

Chapter 8. Security in Computer Networks

8.1 What is Network Security

  • secure communication
    • confidentiality
    • message integrity
    • end-point authentication
    • operational security

8.2 Principles of Cryptography

  • symmetric key cryptography

    • block ciphers
      • DES
      • 3DES
      • AES
    • cipher block chaining

  • public key encryption

    • RSA

8.3 Message Integrity and Digital Signatures

  • cryptographic hash function
    • MD5
    • SHA-1

(skip) (done)