Transport Layer Protocols, Connection Mechanisms, and Flow Control

Transport Layer Fundamentals and Services

• The transport layer is responsible for process-to-process delivery of data across a network.
• Key functions include:     - Demultiplexing of data streams.     - Creating long-lived connections (optional).     - Ensuring reliable, in-order packet delivery (optional).     - Error detection.     - Flow and congestion control (optional).
• Key challenges for this layer involve:     - Detecting and responding to network congestion.     - Balancing fairness among multiple users against high network utilization.
• Layer comparisons in delivery types:     - Node-to-node: Handled by the Data Link Layer.     - Host-to-host: Handled by the Network Layer.     - Process-to-process: Handled by the Transport Layer.
• Comparison of addressing at different layers:     - Data Link Layer: Uses MAC addresses to select one specific node among several on a local segment.     - Network Layer: Uses IP addresses to select one host among millions on the internet.     - Transport Layer: Uses transport layer addresses called Port Numbers to identify specific processes running on a host.         - Destination port numbers are used for delivery.         - Source port numbers are used for the reply.

The Case for Multiplexing and Demultiplexing

• Multiplexing Necessity: Datagram networks (like the internet) do not have circuits or pre-established connections. Since clients run multiple applications simultaneously, the network must identify which application should receive a specific packet.
• IP Header Field: The IP header contains an 8-bit "protocol" field, which allows for up to $256$ concurrent streams. The Transport Layer is inserted to handle complex demultiplexing beyond this limit.
• Demultiplexing Mechanics:     - Traffic endpoints are uniquely identified by a 4-tuple: <src_ip, src_port, dest_ip, dest_port>.     - Each application on a host uses a unique port.     - Multiple applications share the same underlying network connection.     - Server applications use these identifiers to communicate with multiple clients simultaneously.
• Layering Interaction:     - The Transport Layer is theoretically the lowest level end-to-end protocol.     - Transport headers are typically only read by the source and destination hosts.     - Intermediate routers view the transport header merely as part of the data payload.

User Datagram Protocol (UDP)

• Characteristics: Simple, connectionless datagram service.
• Socket Type: In C sockets, it is identified as SOCK_DGRAM.
• Port Numbers: Uses 16-bit identifiers, allowing for $65535$ possible ports. Port $0$ is considered invalid.
• Checksums: Used for error detection to identify corrupted packets. However, UDP does not detect dropped, duplicated, or reordered packets.
• Segment Fields:     - Source Port (16 bits)     - Destination Port (16 bits)     - Payload Length     - Checksum

Transmission Control Protocol (TCP)

• Characteristics: Provides reliable, in-order, bi-directional byte streams.
• Key Features:     - Virtual circuits (logical connections).     - Flow control to prevent receiver overflow.     - Congestion control to approximate fairness in the network.
• TCP Segment Structure:     - Source and Destination Ports: Identifying endpoints.     - Sequence Number (32 bits): Counts bytes in the byte stream (not segments).     - Acknowledgement Number (32 bits): The sequence number of the next expected byte.     - Header Length: Length of the TCP header.     - Flags: 1-bit each (URG, ACK, PSH, RST, SYN, FIN, CWR, ECE).         - SYN: Used for connection setup synchronization.         - ACK: Indicates the acknowledgement field is valid.         - FIN: Used to terminate a connection.         - RST: Reset the connection.     - Receive Window: Used for flow control; indicates the number of bytes the receiver is currently willing to accept.     - Checksum: Error detection for header and payload.     - Urgent Pointer: Points to urgent data.     - Options: Variable length.     - Application Data: Variable length data from the TCP socket.

Connection Management and the Three-Way Handshake

• Importance of Connection Setup: Necessary to establish state on both hosts, primarily initial sequence numbers. These numbers count sent bytes and are chosen randomly to prevent connection confusion from the same host.
• The Three-Way Handshake Process:     - Step 1 (Client): Sends SYN <SeqC, 0>. Client state moves to SYN_SENT.     - Step 2 (Server): Sends SYN/ACK <SeqS, SeqC+1>. Server state moves to SYN_RCVD. This ACKs the client's sequence number and provides its own.     - Step 3 (Client): Sends ACK <SeqC+1, SeqS+1>. Both enter ESTABLISHED state. This segment may contain actual data.
• Why 2-Way Handshake Fails:     - Variable network delays and message reordering.     - Retransmitted connection requests due to loss can lead to half-open connections or duplicate data acceptance.
• Connection Issues:     - SYN Flood: A denial of service (DoS) attack where the server allocates state for every SYN received but never receives the final ACK.     - Solution: SYN cookies.
• Connection Tear Down:     - Either side can initiate.     - Uses a FIN segment.     - Supports "half-open" connections where one side stops sending but continues receiving.     - The final FIN is acknowledged by Seq + 1.

Sequence Number Space and Bidirectional Communication

• Byte Stream Abstraction: TCP treats data as a continuous stream of bytes, each numbered with a 32-bit value that wraps around.
• MSS (Maximum Segment Size): Sets limits on segment size to avoid fragmentation across the network.
• Bidirectional Data Exchange Example:     - Host A to Host B: User types ‘C’: Seq=42, ACK=79, data = ‘C’.     - Host B to Host A: ACK receipt and echo back ‘C’: Seq=79, ACK=43, data = ‘C’.     - Host A to Host B: ACK the echo: Seq=43, ACK=80.
• Piggybacking: Data and ACKs can be sent in the same packet.

Flow Control and the Sliding Window

• Definition: Procedures to restrict the amount of data a sender can transmit before waiting for an acknowledgement. This prevents the sender from overwhelming the receiver's buffer.
• The Problem: If the network delivers data faster than the application layer removes it from the socket buffer, the buffer overflows.
• Mechanism: Sliding Window:     - The receiver provides an Advertised Window (rwnd) in the TCP header.     - For a window size $n$, the sender can transmit $n$ bytes without an ACK.     - The window "slides" forward as ACKs are received.     - ACK Clocking:         - Short RTT → quick ACKs → window slides quickly → high throughput.         - Long RTT → slow ACKs → window slides slowly → low throughput.

Sliding Window Analysis and Network Efficiency

• Scenario for Calculation:     - Bandwidth = $1 \text{ Mbps} = 1,000,000 \text{ bits/sec}$     - RTT = $100 \text{ ms} = 0.1 \text{ sec}$     - Packet size = $1000 \text{ bytes} = 8000 \text{ bits}$
• Step 1: Bandwidth-Delay Product (BDP):     - Determines how much data should be in transit to keep the pipe full.     - $BDP = 1,000,000 \times 0.1 = 100,000 \text{ bits}$
• Step 2: Convert to Packets:     - $\frac{100,000 \text{ bits}}{8000 \text{ bits/packet}} = 12.5 \text{ packets}$
• Step 3: Efficiency Results:     - If window size $≥ 13$: Full link utilization (Throughput = $1 \text{ Mbps}$).     - If window size < $BDP$: Pipe is underutilized.     - Stop-and-Wait (Window = 1):         - $\text{Throughput} = \frac{8000 \text{ bits}}{0.1 \text{ sec}} = 80,000 \text{ bps}$ ($8\%$ efficiency).

Acknowledgement Strategies

• Cumulative ACK: An ACK for sequence $n$ implies receipt of all bytes up to $n-1$. "Next expected is $n$."
• Negative ACKs (NACKs): Explicitly indicate which packet did not arrive.
• Selective ACKs (SACKs): Indicate which packets have arrived, even if they are out of order (e.g., if packets 1, 3, and 4 arrive, SACK specifies 3 and 4 were received despite 2 being lost).