Computer Networks

Sheet 1

Addressing & Subnetting

2 formulas

Core concepts

• Class A: 1-126 (/8); Class B: 128-191 (/16); Class C: 192-223 (/24)
• Subnet mask for /n = n ones then (32-n) zeros

Key formulas

Hosts in a /n Subnet ☆

\[ 2^{32-n}-2 \]

subtract network + broadcast

Subnets from Borrowed Bits ☆

\[ 2^k \]

borrowing k bits

Test your recall

Can you recall the hosts in a /n subnet formula?

Reveal formula

\[ 2^{32-n}-2 \]

GATE traps 4 traps

Don't forget the −2

The '-2' in host-count formulas (network address + broadcast address) is frequently forgotten -- and some questions explicitly allow all-zeros/all-ones subnets, so read the problem statement carefully before subtracting.

Classful addressing is mostly historical

Class-based addressing (A/B/C) is essentially obsolete in real networks (replaced by CIDR) but still appears in GATE numericals -- don't assume a class-default mask applies if an explicit subnet mask is given.

Subnetting borrows from the host side

Subnetting increases network/subnet bits and decreases host bits -- a common sign error is subtracting from the wrong end when recomputing the mask.

Supernetting reverses subnetting

Supernetting (route aggregation) combines multiple smaller blocks into one larger prefix -- the reverse of subnetting; occasionally tested by asking you to aggregate several /24s.

Sheet 2

Transport Layer (TCP/UDP)

3 formulas

Core concepts

• TCP header: 20 bytes minimum; UDP header: 8 bytes fixed

Key formulas

Throughput (Unsaturated) ☆

\[ \frac{\text{Window size}}{RTT} \]

Bandwidth-Delay Product ☆

\[ \text{Bandwidth} \times RTT \]

Stop-and-Wait Efficiency ☆

\[ \frac{T_t}{T_t+2T_{prop}} \]

Test your recall

Can you recall the throughput (unsaturated) formula?

Reveal formula

\[ \frac{\text{Window size}}{RTT} \]

GATE traps 4 traps

UDP still has a checksum

UDP is connectionless with no reliability, ordering, or congestion control -- but it DOES still have a checksum for error detection; assuming UDP has zero error detection at all is a common overreach.

Stop-and-wait struggles on long fat networks

Stop-and-wait efficiency collapses when propagation delay is large relative to transmission time (long fat networks) -- this is exactly why sliding window protocols exist, a frequently tested 'why' question.

Handshake ≠ teardown sequence

TCP's three-way handshake (SYN, SYN-ACK, ACK) is for connection SETUP; a separate four-step exchange (FIN, ACK, FIN, ACK) handles graceful TEARDOWN -- mixing these two sequences up is a common mistake in step-numbered questions.

Window size has two ceilings

Window size is bounded by BOTH the sender's buffer and the sequence-number space (2^(sequence bits)) -- numericals sometimes test which constraint is actually the binding one.

Sheet 3

Data Link Layer & Routing

2 formulas

Core concepts

• CRC: append (n-1) zero bits for an n-bit generator, divide, remainder = appended CRC bits
• Distance-vector (Bellman-Ford based) vs Link-state (Dijkstra based) routing

Key formulas

Hamming Code Parity Bits ☆

\[ 2^r \ge m+r+1 \]

Sliding Window Bounds ☆

\[ \text{GBN} \le 2^k{-}1 \qquad \text{SR} \le 2^{k-1} \]

Test your recall

Can you recall the hamming code parity bits formula?

Reveal formula

\[ 2^r \ge m+r+1 \]

GATE traps 4 traps

GBN discards, SR buffers

Go-Back-N discards ALL out-of-order frames at the receiver; Selective Repeat buffers out-of-order frames individually -- this exact behavioral difference is what most GBN-vs-SR questions hinge on.

SR's window is half of GBN's

Selective Repeat's window bound is HALF that of Go-Back-N for the same sequence-number bits (2^(k-1) vs 2^k - 1) -- using the wrong formula for the wrong protocol is the single most common error on this topic.

Distance-vector can count to infinity

Distance-vector protocols suffer from 'count-to-infinity' on link failure -- split horizon and route poisoning only partially mitigate it, they don't fully fix it.

Hamming corrects 1, only detects 2

Hamming code corrects single-bit errors and can DETECT (not correct) double-bit errors -- conflating detection with correction capability is a common conceptual slip.

Sheet 4

Application Layer & Ports

1 formula

Core concepts

• DNS resolves names top-down: Root -> TLD (.com/.org) -> Authoritative name server
• HTTP is stateless -- each request is independent; cookies/sessions are the application-level workaround for tracking state

Key formulas

Well-Known Port Range ☆

\[ 0 \text{ to } 1023 \]

registered: 1024–49151, dynamic/private: 49152–65535

Test your recall

Can you recall the well-known port range formula?

Reveal formula

\[ 0 \text{ to } 1023 \]

Reference table

Protocol	Port	Transport
HTTP	80	TCP
HTTPS	443	TCP
FTP (control)	21	TCP
FTP (data)	20	TCP
SSH	22	TCP
Telnet	23	TCP
SMTP	25	TCP
DNS	53	UDP / TCP
DHCP	67, 68	UDP
POP3	110	TCP
IMAP	143	TCP
SNMP	161	UDP

GATE traps 3 traps

DNS isn't strictly UDP

DNS uses UDP for regular queries (fast, best-effort, retried by the resolver on loss) but falls back to TCP for zone transfers and responses larger than 512 bytes -- assuming DNS is UDP-only is a common oversimplification GATE exploits.

FTP uses two connections

FTP uses TWO separate connections -- a control connection on port 21 that stays open for the whole session, and a data connection on port 20 opened per transfer -- most other application protocols (HTTP, SMTP) use just one.

Telnet and FTP are plaintext

Telnet and FTP send credentials in PLAINTEXT (no encryption) -- SSH (port 22) and SFTP are their secure replacements; GATE sometimes tests which protocols are inherently insecure.