T4b

Transport Layer: TCP/UDP Introduction

Course Information

• Instructor: Dr. Markus Buchhorn
• Email: markus.buchhorn@anu.edu.au

Layer Stack

• The transport layer resides above the network layer (IP) and below the application layer.
• The order of layers is as follows:
  • Application
  • Transport
  • Network (IPv4, IPv6)
  • Link (Ethernet, WiFi)
  • Physical (Cables, Space and Bits)
• Each layer deals with different units of data:
  • Messages (Application Layer)
  • Segments/Datagrams (Transport Layer)
  • Packets (Network Layer)
  • Frames (Link Layer)
  • Bits (Physical Layer)

Focus on Applications and Transport Layer

• The network layer's complexities are generally abstracted away from applications.
• Applications are mostly unaware of the underlying network links unless performance issues arise.

Transport Layer Functionality

• The transport layer uses IP packets for transmission.
• It provides functionalities such as reliability, performance optimization, security measures to applications, operating on top of unreliable IP.
• Network devices (routers) operate at the network layer and do not interfere with the transport layer's payload, focusing on the 'envelope' of the message.
• The transport layer ensures host-to-host communication.

Client/Server Model

• Servers offer services.
• Clients connect to servers, send requests, and receive replies.
• Servers can handle multiple clients simultaneously.
• Note that alternative models exist.

Transport Services: Reliability and Communication Types

• Common application needs include reliability and different communication types.
• Reliability:
  • Reliable: Ensures bit-perfect arrival of data by repairing packet loss, mis-ordering, and errors, which may introduce delay.
  • Unreliable: Prioritizes performance and simplicity over immediate perfection.
• Communication Types:
  • Messages: Self-contained commands and responses.
  • Byte-stream: Continuous flow of bytes, chunked into segments.

TCP vs. UDP

• UDP (User Datagram Protocol):
  • Unreliable.
  • Messages (Datagrams).
• TCP (Transmission Control Protocol):
  • Reliable.
  • Byte-stream (Streams).
• It is possible to build reliable messages on top of TCP.
• Using unreliable byte-streams would resemble UDP.
• Protocol Numbers:
  • TCP is IP Protocol 6.
  • UDP is IP Protocol 17.
  • ICMP is 1, IGMP is 2, IPv6 encapsulation is 41, and there are 130+ more.

Comparison of UDP and TCP

IP Multicast with UDP

• UDP is suitable for IP multicast due to its connectionless nature and tolerance for replica packets, which is useful in time-sensitive applications.

Ports

• Ports enable process-to-process communication on hosts, enhancing the host-to-host communication provided by IP.
• A port is a 16-bit identifier for a process on a host, associated with a network interface.

Well-Known Ports

• Ports below 1024 require elevated privileges to open.
• Examples:
  • 20, 21: ftp (File Transfer Protocol)
  • 22: ssh (Secure Shell)
  • 25: smtp (Simple Mail Transfer Protocol) - outbound email
  • 80: http (Web)
  • 110: pop3 (Post Office Protocol version 3) - inbound email
  • 143: imap (Internet Message Access Protocol) - inbound email
  • 443: https (Secure Web)
• Application addresses are typically written as IP:port, e.g., 150.203.56.90:80.
• For an extensive list, refer to: https://www.iana.org/assignments/service-names-port-numbers/service-names-port-numbers.xhtml

Port Management

• Inetd/xinetd:
  • Manages multiple server services.
  • Listens to registered ports and protocols (tcp, udp).
  • Launches appropriate service on demand.
• Port mapping:
  • Used by services like remote procedure calls or BitTorrent.
  • Listens on a well-known port, accepts connections, and redirects them to a spawned service on another port.
  • Services register with a portmapper.

NAT (Network Address Translation)

• NAT hides multiple devices behind a single public IP address.
• It translates addresses and ports, maintaining a dynamic table for outbound connections and a static table for inbound connections.
• Example:
  • Internal IP 10.0.0.2:80 is translated to 150.203.56.99:7882.
  • Internal IP 10.0.0.4:80 is translated to 150.203.56.99:7884.

Socket Programming

• Socket programming involves using an address, a port, and communication needs.
• Connections are uniquely identified in the operating system by a 5-tuple:
  • Source IP
  • Destination IP
  • Source Port
  • Destination Port
  • Protocol

Socket API Primitives

Connection Sequence

• Server Preparation:
  1. socket()
  2. bind()
  3. listen() / receivefrom()
  4. accept()
• Client Initiation:
  1. socket()
  2. connect()
  3. recv()
  4. send()
  5. recv()
  6. send()
  7. close()
• Finally, the server executes close()
• Calls marked with an asterisk (*) may block.

UDP Details

• UDP adds ports, payload length, and a checksum to IP.
• Its header includes source port, destination port, length, and checksum.

TCP Details

• TCP segments carry chunks of a byte-stream, without preserving message boundaries.
• The sender's transport layer packetizes data on write() calls, with multiple writes forming one packet or vice-versa based on buffering.
• The receiver unpacks the byte-stream, and applications read() the stream in order, without loss.

TCP Segment Format

• The header comprises source port, destination port, sequence number, acknowledgement number, header length, flags (CWR, ECE, URG, ACK, PSH, RST, SYN, FIN), window size, checksum, urgent pointer, and options.
• It includes ports, reliability, performance, acknowledgement.
• Start and stop bits reside in the header, not the payload.

TCP Reliability

• TCP provides a reliable, bidirectional byte-stream using sequence numbers and acknowledgements.
• Sequence numbers:
  • N-bit counter that wraps (e.g., …,253, 254, 255, 0, 1, 2…).
  • Represents a byte count (pointer) in a stream.
  • Can wrap quickly on high-speed links ( $2^{32} = 4$ GB).
  • Timestamps can also be used.
  • Does not start from zero (for security).
• Acknowledgements:
  • Indicate which byte is expected (bytes received).
  • Cumulative ACK.

TCP and Payload Length

• Payload length can be calculated from the increase in Seq #
• Losses are easily detectable.
• TCP is full-duplex, utilizing two simplex paths.
• It Piggybacks control information on data segments in reverse direction, or sends just feedback if no data is available.

TCP Connection Establishment (3-Way Handshake)

• TCP is full-duplex, requiring both paths to start together.
• It Synchronizes Sequence numbers in both directions.
• Connecting: Receiving transport layer checks if anything is listen()ing on that port.
  • If not, it sends a ReSeT.
  • Transport layer ACKnowledges and SYNs its own #.
  • Originator ACKs that SYN/ACK.

TCP Disconnection

• Both sides need to end together, ideally, to flush buffers.
• Disconnecting: One side initiates close(), triggering a FIN(alise).
  • The other side ACKs and FINs too.
  • If FIN is lost, it is resent.

Socket States

Netstat Example

• The netstat -n command shows network connections.
• The output includes protocol, local address, foreign address, and state.
• It displays what's happening on the machine, listing listening ports, established connections, and various states of TCP and UDP connections.

TCP Sliding Windows

• TCP aims for both reliability and throughput.
• It starts with ARQ (Automatic Repeat Request) – stop-and-wait, where a single segment outstanding can be a problem on high delay networks.
• If one-way-delay is 50ms, then round-trip-time (RTT) = $2d = 100$ms.
• A Single segment per RTT = 10 segments/s.
• With a typical packet size of 1000 bytes (10,000 bits), the throughput is approximately 100kb/s.
• Increasing bandwidth alone doesn't improve throughput.

TCP Sliding Windows: Optimizing Throughput

• Allow $W$ bytes to be 'outstanding' (unACKed) per RTT to fill a pipeline/conveyor-belt with segments.
• Set up a 'window' of $W$ bytes.
• $W = 2 * Bandwidth * delay$ (roughly).
• At 100Mb/s, with a one-way-delay of 50ms, $W = 10$Mb (~1MB).
• Assuming 1000-byte segments, W = 1000 segments are out there somewhere!

Sliding Window Approach

• Sender buffers up $W$ segments until they are ACKed.
• If the window isn't full, a segment is sent.
• When a segment is ACKed, the window slides, making space available.

ARQ (Automatic Repeat Request) - If(lost) then:

• Receiver buffers just a single segment.
• If it’s the next one in sequence, ACK it, everyone happy.
• If it’s not, drop it.
• Let sender retransmit what I’m actually waiting for.
• Sender has a single timer. After timeout, resend (all) from (first) ACK-less.
• Really simple, but somewhat inefficient(!)
• Need to understand more about loss and performance behaviours.

TCP Implementations

• There is no single TCP implementation.
• This doesn’t impact the network, only hosts, so you can do what you want… – Maybe, maybe not…
• Many years of various optimisations, experiments, algorithms, … suited to various circumstances.
• And as vulnerabilities have been found and mitigated (and found and …)