Transport Layer Protocol Practice Questions

In Stop-and-Wait, sender transmits for $T_t$ seconds, then waits for ACK which takes $2T_p$ (one propagation delay each way). Total cycle = $T_t + 2T_p$. Efficiency = $T_t / (T_t + 2T_p)$.

A: TRUE - TCP connection establishment uses 3-way handshake: SYN (client), SYN-ACK (server), ACK (client). B: TRUE - the rwnd field is 16 bits → max 65,535 bytes ≈ 64KB. The TCP Window Scale option (RFC 1323) extends this by specifying a scaling factor (up to 2^14), allowing windows up to 1GB for high-speed networks. C: TRUE - TCP connection teardown uses 4-way handshake: FIN (active closer), ACK (passive closer), FIN (passive closer), ACK (active closer). D: TRUE - persist timer fires when rwnd = 0, causing the sender to probe with 1-byte segments to prevent deadlock if the window update is lost.

Transmission time $T_t = \frac{1000}{10^6} = 1$ ms. Propagation delay $T_p = 10$ ms. Efficiency = $\frac{T_t}{T_t + 2 T_p} = \frac{1}{1 + 20} = \frac{1}{21} \approx 0.0476$. In percentage: $0.0476 \times 100 = 4.76\% \approx 4.8\%$.

QUIC (Quick UDP Internet Connections), used in HTTP/3: A: TRUE - QUIC runs over UDP, so head-of-line blocking within one stream doesn't block other streams (unlike TCP where all streams on one connection stall if a packet is lost). B: TRUE - QUIC integrates TLS 1.3, combining transport and crypto handshakes: 0-RTT or 1-RTT connection establishment vs TCP+TLS 1.2's 3-RTT. D: TRUE - QUIC uses connection IDs instead of 4-tuple (src IP, src port, dst IP, dst port), allowing connections to survive IP changes (e.g., mobile handoff). C: FALSE - QUIC implements its own congestion control (similar to CUBIC/BBR), it does not eliminate congestion control.

Well-known ports range from 0 to 1023 and are assigned by IANA for standard services. Port 80 is the well-known port for HTTP. Ports 1024-49151 are registered ports, and 49152-65535 are dynamic/private ports.

Karn's algorithm states that RTT samples from retransmitted segments should not be used for RTT estimation because the acknowledgment could correspond to either the original or retransmitted segment, creating ambiguity. Additionally, the retransmission timer is backed off exponentially after each timeout until an ACK for new data is received.

On a timeout, TCP sets ssthresh = cwnd/2 (half the current congestion window), then resets cwnd to 1 MSS and restarts slow start.

Large propagation delay + high transmission rate reduces utilization.

In TCP flow control, the sender can have at most rwnd bytes of unacknowledged data outstanding at any time. Currently unacknowledged = 4,000 bytes. Remaining send window = rwnd - unacknowledged data = 10,000 - 4,000 = 6,000 bytes. The sender can transmit 6,000 more bytes before it must wait for acknowledgments. This is the essence of the sliding window - as acknowledgments arrive, the window slides forward allowing more data to be sent.

ssthresh (slow start threshold) divides TCP congestion control into two phases: When cwnd < ssthresh: Slow Start - cwnd grows exponentially (doubles per RTT). When cwnd ≥ ssthresh: Congestion Avoidance - cwnd grows linearly (+1 MSS per RTT). When congestion is detected: ssthresh is set to cwnd/2 (half the current window), then cwnd is reset (to 1 on timeout, to ssthresh on 3 dup ACKs). ssthresh is dynamically updated and represents the best estimate of the network's safe capacity based on the most recent congestion event.

Window-limited throughput = $\frac{\text{Window Size} \times 8}{RTT} = \frac{52000 \times 8}{0.08} = \frac{416000}{0.08} = 5,200,000$ bps = $5.20$ Mbps. Since $5.20$ Mbps < $10$ Mbps (link bandwidth), the receiver window is the bottleneck.

In congestion avoidance, cwnd increases by 1 MSS per RTT (linear increase), compared to slow start where it doubles (exponential). This is the AIMD (Additive Increase Multiplicative Decrease) behavior.

The Transport Layer provides end-to-end communication services including segmentation of data, flow control to prevent overwhelming the receiver, and error control to ensure reliable delivery. Routing is a Network Layer function, physical addressing is Data Link/Physical layer, and logical addressing is Network Layer.

In slow start, the congestion window (cwnd) doubles every RTT (increases by 1 MSS per acknowledged segment), resulting in exponential growth until the slow start threshold (ssthresh) is reached.

ECN (Explicit Congestion Notification, RFC 3168) allows routers to signal congestion without dropping packets. Mechanism: (1) ECN-capable endpoints set the ECT (ECN-Capable Transport) bit in IP headers. (2) A congested router sets the CE (Congestion Experienced) bit in the IP header instead of dropping the packet. (3) The receiver echoes this CE signal to the sender using the ECE (ECN-Echo) bit in the TCP header. (4) The sender reduces cwnd as if a loss occurred and sets CWR (Congestion Window Reduced) bit. ECN avoids the latency penalty of packet loss while still responding to congestion.

In Go-Back-N, the receiver has no buffer for out-of-order frames. When frame 3 is lost: frames 4, 5, 6 arrive out of order and are discarded by the GBN receiver (which expects frame 3 next). For each out-of-order frame received, the receiver sends a duplicate ACK for the last correctly received in-order frame (ACK 2, meaning frames 0,1,2 received). The sender, after timeout or receiving n dup-ACKs, retransmits ALL frames starting from frame 3: retransmits 3, 4, 5, 6. This is the 'go back N' behavior - wasteful but simple.

The sender's effective window is $\min(cwnd, rwnd) = \min(24, 16) = 16$ KB. The sender can send up to the minimum of congestion window and receiver window. Since $16$ KB is already a multiple of MSS ($2$ KB), the answer is $16$ KB.

Maximum TCP throughput = Window Size / RTT = $W/R$ bytes per second. This is the fundamental bandwidth-delay product constraint on TCP throughput.