Network Congestion and Delays

Study Notes

Two senders and two receivers communicate with each other over a shared output link with infinite buffers.
The output link capacity is R.
Each sender can achieve a maximum throughput of R/2 due to shared resources.
Large delays can occur as the arrival rate approaches the output link capacity (R).

Two hosts communicate with a router that has finite buffers.
Retransmissions occur when packets time out.
The original data rate is 'lin', and the data rate including retransmissions is 'l'in.
The output link capacity is R.
The router can act optimally when perfectly aware of its available buffer space, limiting the sender's output to its available space.
The sender can achieve a throughput of R/2 by sending only when buffer space is available.
Packets can be lost due to full buffers, requiring retransmission.
The sender can achieve a rate of R/2 asymptotically where some packets are lost and some are retransmitted because the goodput remains R/2.
Packets can be duplicated due to premature timeouts, where a sender retransmits before confirmation of loss.

TCP segments include a sequence number to track data flow.
TCP segments are reliably delivered, ensuring data reaches the destination in the order they are sent.
TCP segments include a checksum for data integrity.

The TCP 3-way handshake is a process for establishing a connection between two hosts.
The handshake involves three message exchanges:
- Client sends a SYN (synchronization) packet to establish a connection.
- The server sends a SYN-ACK packet, confirming receipt of the SYN packet and acknowledging the connection attempt.
- The client acknowledges the SYN-ACK packet, completing the connection establishment.

Client and server independently close their respective ends of a connection.
This is done by both sending a FIN (finish) segment with the FIN bit set to 1.
Each party acknowledges the received FIN segment in a separate ACK segment.
FIN and ACK can be combined in a single segment if they are received and sent simultaneously.
Closing a connection progresses through a series of states to ensure a clean shutdown.

Congestion occurs when networks are overloaded with traffic.
Congestion control mechanisms aim to prevent or minimize congestion by managing traffic flow.

End-to-end congestion control relies on end systems to infer congestion based on loss or delay, without explicit feedback from the network.
Network-assisted congestion control involves routers providing explicit feedback to end systems, including congestion indicators or rate control information.

ATM (Asynchronous Transfer Mode) ABR (Available Bit Rate) is an elastic service where senders adjust their rate based on network conditions.
Resource Management (RM) cells are used in conjunction with data cells.
RM cells contain bits set by switches to indicate congestion levels:
- NI (No Increase) bit: Mild congestion, sender should not increase rate.
- CI (Congestion Indication) bit: Severe congestion, sender should reduce rate.
RM cells are returned to senders by receivers to provide congestion feedback.

TCP uses an approach of additive increase and multiplicative decrease to automatically adjust its transmission rate (window size) in response to network conditions.
Additive increase: The window size is increased by 1 MSS (Maximum Segment Size) every Round Trip Time (RTT) until loss is detected.
Multiplicative decrease: The window size is cut in half after a loss event.