Data-Link Technologies and High-Speed Internetworking

Introduction and Context

Data-link technologies provide the mechanisms by which directly connected network nodes exchange frames over a physical or logical link. At postgraduate level, their importance extends beyond frame formatting: data-link design determines utilization, latency, resilience, fault domains, quality of service, and the scalability of internetworks. Modern carrier, enterprise, cloud, and industrial networks combine Ethernet switching, VLAN segmentation, link aggregation, tunnelling, MPLS, and IPv4/IPv6 transition mechanisms.

The data-link layer is commonly associated with Layer 2 of the OSI model, although practical technologies often cross strict layer boundaries. Ethernet switches, for example, make forwarding decisions using Layer 2 addresses but may also classify traffic using IP, transport ports, or application policy. Similarly, MPLS labels are inserted between Layer 2 and Layer 3 headers, while carrier Ethernet delivers Ethernet services across provider infrastructure.

Learning Objectives

After completing this chapter, the learner should be able to:

1. Explain framing, error detection, error correction, flow control, and ARQ mechanisms.

2. Compare MAC protocols using collision probability, efficiency, fairness, and delay.

3. Analyze Ethernet switching, VLANs, trunks, spanning-tree operation, and link aggregation.

4. Explain MPLS forwarding, control-plane operation, traffic engineering, and scalability constraints.

5. Distinguish carrier Ethernet services from conventional LAN Ethernet.

6. Explain tunnelling and the operational issues of IPv4/IPv6 interoperability.

7. Select appropriate technologies for high-speed enterprise and carrier-network designs.

1. Framing and Reliable Data Transfer

1.1 Framing functions

A frame is the data-link-layer unit exchanged between adjacent nodes. Framing performs several functions:

• identifies the beginning and end of a transmission;
• carries addressing and control information;
• supports error detection;
• separates payloads belonging to different higher-layer packets; and
• permits synchronization and, in some systems, multiplexing.

Common framing methods include character count, byte-oriented sentinel framing, bit-oriented flag framing, and physical-layer coding violations. Character-count framing is simple but vulnerable to loss of synchronization if the count field is corrupted. Sentinel methods use a reserved flag and an escape mechanism. In bit-oriented protocols such as HDLC, a flag pattern such as 01111110 is protected by bit stuffing: after five consecutive one bits in the data, the sender inserts a zero, which the receiver removes.

Ethernet uses a preamble and start-frame delimiter for receiver synchronization, followed by destination and source MAC addresses, an EtherType or length field, payload, and frame check sequence (FCS). The Ethernet payload has a minimum size in traditional shared-medium Ethernet, historically supporting collision detection and minimum transmission time. Modern switched full-duplex Ethernet does not normally experience collisions, but the frame format remains compatible with established standards.

1.2 Error detection

Error detection does not recover a corrupted frame; it determines whether retransmission or higher-layer action is necessary. Parity detects some odd numbers of bit errors but has weak detection capability. A checksum treats data words arithmetically and is inexpensive, while a cyclic redundancy check (CRC) treats the frame as a polynomial over GF(2).

For a generator polynomial G(x), a transmitter appends r check bits to a message polynomial M(x), producing a codeword divisible by G(x). The receiver divides the received polynomial by G(x); a non-zero remainder indicates an error. A well-selected CRC detects all burst errors of length at most r and detects many longer bursts, but no finite CRC guarantees detection of every possible error pattern.

CRC performance depends on the generator polynomial and frame length. It should therefore not be described simply as “error correction.” Ethernet’s FCS is principally an error-detection mechanism; corrupted frames are discarded, and recovery is normally delegated to TCP, a wireless ARQ mechanism, or another protocol.

1.3 Flow control

Flow control prevents a fast sender from overrunning a receiver. Stop-and-wait allows one outstanding frame and requires an acknowledgment before proceeding. Sliding-window protocols allow multiple unacknowledged frames and use sequence numbers to identify them.

If a link has rate R bits/s, round-trip time RTT, and frame size L bits, the stop-and-wait utilization under negligible processing time is approximately

U = L/(L + R·RTT).

This expression assumes one frame transmission, an acknowledgment exchange represented by RTT, no losses, and no pipelining. It shows why stop-and-wait performs poorly on high-bandwidth-delay-product paths. For a sliding window of W frames, a useful idealized condition for full utilization is

W·L ≥ R·RTT,

or, equivalently, W should be at least the bandwidth-delay product divided by frame size. Protocol overhead, acknowledgment transmission time, and loss recovery reduce actual utilization.

2. ARQ and Probabilistic Performance

Automatic Repeat reQuest (ARQ) combines error detection, acknowledgments, sequence numbers, timers, and retransmission. Its principal forms are stop-and-wait ARQ, go-back-N (GBN), and selective repeat (SR).

2.1 Stop-and-wait ARQ

The sender transmits one frame and waits for an ACK. A negative acknowledgment or timeout causes retransmission. Sequence numbers distinguish a delayed duplicate from a new frame. Stop-and-wait is simple but wastes capacity when propagation delay is large.

Assume an independent frame-error probability p, immediate successful acknowledgment when the data frame is correct, and unlimited retransmissions. The number of transmissions N required for success is geometrically distributed:

P(N = k) = (1 − p)p^(k−1), k ≥ 1.

Therefore,

E[N] = 1/(1 − p).

If a successful data exchange consumes a cycle time T and retransmission attempts have the same duration, the expected time per successfully delivered frame is approximately T/(1 − p). This is a simplified model: acknowledgment losses, correlated errors, timer overhead, and variable frame lengths require a more detailed model.

For example, with p = 0.1, E[N] = 1/0.9 = 1.111 transmissions per successful frame. Thus, retransmission work increases by approximately 11.1% relative to an error-free link. With p = 0.2, E[N] = 1.25, a 25% increase. The result assumes independent attempts; burst errors can make actual performance substantially worse.

2.2 Go-back-N

GBN permits a window of W frames but uses cumulative acknowledgments. If frame i is lost, the receiver may discard subsequent out-of-order frames, and the sender retransmits frame i and all later outstanding frames. GBN has modest receiver complexity but can waste bandwidth on links with non-negligible error probability.

A rough efficiency estimate under ideal acknowledgments is

η ≈ min(1, W·T_f/T_cycle)·(1 − p),

where T_f is frame-transmission time and p is the probability that an attempted frame causes a recovery event. This is not a universal GBN formula: the exact result depends on window size, timeout behavior, cumulative acknowledgment timing, and whether p represents independent frame errors or complete window failures.

2.3 Selective repeat

SR acknowledges frames individually and buffers out-of-order frames. Only missing or damaged frames are retransmitted. It is more efficient than GBN on high-error or long-delay paths but requires receiver memory, more sequence-number state, and careful window constraints. To avoid ambiguity between old and new frames, the sequence-number space must be sufficiently large; in the common bounded-window design, the sequence space is at least twice the window size.

At postgraduate design level, ARQ selection should consider goodput rather than nominal link rate. Goodput is useful payload delivered per unit time. SR generally provides higher goodput under sparse independent losses, whereas GBN may be competitive when errors are rare and implementation simplicity is valuable. FEC may outperform ARQ when retransmission latency is unacceptable, such as real-time media or satellite links.

3. Medium Access Control

The MAC sublayer determines how multiple stations share a medium. Protocols include random access, controlled access, and channelization.

3.1 Random access

In pure ALOHA, a frame can be transmitted at any time. If the frame transmission time is T, a frame is vulnerable to transmissions beginning during an interval of length 2T. Under the classical Poisson model with offered load G frames per frame time, throughput is

S = G e^(−2G),

with maximum S = 1/(2e) ≈ 0.184 at G = 0.5. The model assumes independent transmission attempts, a single shared channel, and idealized immediate collision feedback.

Slotted ALOHA restricts transmissions to slot boundaries. Its vulnerable period is one slot, giving

S = G e^(−G),

with maximum throughput 1/e ≈ 0.368 at G = 1. These are theoretical maxima, not guaranteed operational performance.

Carrier Sense Multiple Access (CSMA) reduces collisions by sensing before transmission. In CSMA/CD, a station detects collisions during transmission and applies binary exponential backoff. CSMA/CD was important in shared coaxial Ethernet but is not used for ordinary full-duplex switched Ethernet, where each link is point-to-point and collision-free.

Wireless networks use CSMA/CA because a station generally cannot detect a collision reliably while transmitting. Interframe spaces, random backoff, acknowledgments, and optional RTS/CTS reduce collision and hidden-terminal effects. However, sensing and backoff introduce overhead and do not guarantee deterministic delay.

3.2 MAC efficiency and design interpretation

A useful first-order efficiency model for a contention system is

η = useful successful transmission time / (successful time + collision time + idle/backoff time + protocol overhead).

This identity is definitional rather than a universal closed-form equation. It emphasizes that increasing offered load may increase throughput initially but eventually increases collisions and delay. A high-speed network should therefore be evaluated using throughput, delay distribution, collision or retransmission probability, fairness, energy use, and behavior under overload.

Channelization methods such as time-division, frequency-division, and code-division allocate resources more predictably but may waste capacity when traffic is bursty. Scheduling methods can provide quality-of-service guarantees, at the cost of coordination and implementation complexity.

4. Ethernet Switching and Segmentation

4.1 Learning and forwarding

An Ethernet switch maintains a forwarding or switching database. It learns a source MAC address from the frame’s incoming port and records an association such as (source MAC, VLAN, port, aging timer). When forwarding, it looks up the destination MAC address. If the destination is known, the frame is sent only through the associated port; if unknown, the frame is flooded within the VLAN. Broadcast and relevant multicast frames are also replicated within the broadcast domain.

Learning from source addresses is fundamental: the destination address has not yet proved where the frame entered the network. Entries age out because devices move, links change, and stale associations must not persist indefinitely.

4.2 Loops and spanning tree

Layer-2 loops can cause broadcast storms, duplicate frames, and unstable MAC learning. Spanning Tree Protocol (STP) creates a loop-free logical topology by selecting a root bridge and blocking redundant paths. Rapid STP reduces convergence time, while multiple-instance approaches map VLAN groups to spanning-tree instances.

Spanning tree trades active-path redundancy for loop prevention. Modern designs often use link aggregation, routed access layers, or multi-chassis link aggregation to obtain redundancy without depending exclusively on a single tree. Convergence remains a critical concern for industrial control, financial trading, and data-centre applications.

4.3 VLANs and trunks

A VLAN creates a separate Layer-2 broadcast domain over shared switching infrastructure. IEEE 802.1Q tagging inserts a VLAN identifier in trunk frames. VLANs support segmentation, policy boundaries, traffic engineering, and operational organization, but a VLAN is not automatically a security boundary. Misconfiguration, native-VLAN inconsistencies, trunk exposure, and insufficient inter-VLAN controls can permit unauthorized access.

Inter-VLAN communication requires a Layer-3 function, either a router or a multilayer switch. VLAN planning should avoid excessive broadcast domains, preserve address and policy clarity, and use explicit trunk allow-lists. For large-scale data centres, VLAN identifiers alone may not scale across all tenants; overlays such as VXLAN commonly extend segmentation using a larger virtual-network identifier.

4.4 Link aggregation

Link aggregation combines multiple physical links into one logical bundle. IEEE 802.1AX Link Aggregation Control Protocol (LACP) negotiates member links and detects certain configuration mismatches. Aggregation improves total capacity and resilience, but a single flow is normally assigned to one member using a hash over fields such as MAC addresses, IP addresses, or transport ports. Consequently, four 10-Gbit/s links do not necessarily provide 40 Gbit/s to one TCP flow.

Design analysis must account for hash imbalance, failure detection time, peer support, oversubscription, and whether traffic consists of many sufficiently independent flows.

5. MPLS and High-Speed Internetworking

5.1 Forwarding model

Multiprotocol Label Switching (MPLS) forwards packets using short labels rather than performing a full longest-prefix lookup at every transit hop. A label-switched path (LSP) is a logical forwarding path. Provider-edge routers classify packets into forwarding equivalence classes (FECs), impose labels, and transit routers swap labels. At the egress, labels are removed or the penultimate hop may perform penultimate-hop popping.

An MPLS label stack can support hierarchical services. The outer label may identify a provider transport LSP, while an inner label identifies a VPN or service. Label operations include push, swap, and pop. MPLS is not encryption; confidentiality requires an appropriate security mechanism such as IPsec or a trusted protected transport.

5.2 Control plane and scalability

MPLS data-plane forwarding is only useful when labels and paths are distributed consistently. Label Distribution Protocol (LDP) commonly follows the interior routing topology to establish transport labels. Resource Reservation Protocol–Traffic Engineering (RSVP-TE) can establish explicitly constrained paths and reserve or signal resources. Segment Routing can encode path instructions using segment identifiers, reducing dependence on per-LSP signaling state in some architectures.

Scalability must be analyzed in several dimensions:

• Forwarding-table state: transit nodes need label entries, while edge nodes may maintain large VPN or service state.
• Control-plane churn: topology changes trigger routing recalculation and label update activity.
• VPN scale: provider-edge devices may maintain per-customer routes and labels; route reflectors reduce peering complexity but do not eliminate total state.
• Operational complexity: multiple protocols, label spaces, policies, and failure modes increase troubleshooting difficulty.
• Path-state distribution: explicit traffic engineering can improve utilization but may create substantial path and reservation state.

A fully meshed design among n edge routers requires n(n−1)/2 pairwise sessions, which grows quadratically. Route reflection and hierarchical architectures reduce the number of direct sessions, but they introduce control-plane concentration and possible suboptimal routing. Thus MPLS scalability is not merely a question of label length; it is a systems problem involving state, computation, signaling, memory, convergence, and operations.

5.3 MPLS traffic engineering

Traffic engineering attempts to map demands onto network capacity while meeting constraints such as bandwidth, latency, affinity, and protection requirements. A shortest-path routing protocol may leave some links congested while others are lightly used. Explicit LSPs can distribute traffic more effectively, but optimization is based on traffic estimates that may be uncertain or change rapidly. Fast reroute improves availability but consumes additional protection capacity and increases policy complexity.

6. Carrier Ethernet

Carrier Ethernet extends Ethernet services across metropolitan and wide-area provider networks with operational, administrative, and performance requirements beyond a conventional LAN. Important service models include:

• E-Line: point-to-point Ethernet connectivity;
• E-LAN: multipoint-to-multipoint connectivity; and
• E-Tree: rooted multipoint service, often separating a hub from spokes.

Provider networks may use VLAN stacking, MPLS, Provider Backbone Bridging, or other encapsulation mechanisms to separate customer traffic and preserve service identity. Service-level agreements can specify committed information rate, excess information rate, frame delay, delay variation, frame loss, availability, and restoration time.

Carrier Ethernet requires demarcation, Operations, Administration, and Maintenance (OAM), fault isolation, protection switching, and accurate performance measurement. A customer-facing Ethernet port does not imply that the provider network is a single broadcast domain.

7. Tunnelling and IPv4/IPv6 Interoperability

Tunnelling encapsulates one protocol inside another to create a logical path across an intermediate network. Examples include GRE, IP-in-IP, IPsec tunnel mode, and IPv6-over-IPv4 mechanisms. Encapsulation adds overhead and may reduce the effective path MTU. It can also obscure traffic from intermediate quality-of-service, filtering, or troubleshooting functions.

7.1 MTU and fragmentation

If a path MTU is 1500 bytes and an IPv4 packet has a 20-byte header with no options, the maximum IPv4 payload carried in one packet is

1500 − 20 = 1480 bytes.

For an IPv4 packet with a 20-byte header and a TCP segment whose TCP header is 20 bytes, the maximum TCP data is

1500 − 20 − 20 = 1460 bytes.

For IPv6, the base header is normally 40 bytes, so the maximum transport payload in a 1500-byte packet, ignoring extension headers, is 1500 − 40 = 1460 bytes. IPv6 routers do not fragment transit packets; the source uses Path MTU Discovery, while fragmentation, if required, is performed by the source using an extension header. Tunnels reduce the effective inner MTU by the outer-header and tunnel-overhead size. Implementations should therefore use accurate MTU discovery, TCP Maximum Segment Size adjustment where appropriate, and careful handling of ICMP or ICMPv6 Packet Too Big messages.

7.2 Transition mechanisms

Dual stack operates IPv4 and IPv6 simultaneously and is conceptually direct, but it requires addressing, routing, security, monitoring, and operational support for both protocols. Tunnelling carries one protocol through infrastructure that supports another, but adds overhead and dependency on tunnel endpoints. Translation mechanisms such as NAT64/DNS64 allow IPv6-only clients to reach IPv4 services, but can affect application transparency, logging, troubleshooting, and protocols carrying literal IP addresses.

IPv4/IPv6 interoperability is therefore an architectural governance issue, not only a packet-format issue. Address planning, policy parity, DNS behavior, security inspection, asset inventory, and incident response must cover both protocol families.

Applied Case: Illustrative University Research Network

Background

A university operates laboratories, student residences, administrative systems, and a high-performance computing cluster. The campus has redundant 10-Gbit/s uplinks, legacy IPv4 applications, growing IPv6 deployment, and a provider Ethernet service connecting two sites.

Problem

Broadcast growth affects residence networks, a single uplink becomes congested during research transfers, and the two sites require tenant separation and predictable recovery. Some applications fail when tunnel overhead causes packets to exceed the path MTU.

Analysis

The engineering team maps broadcast domains, measures flow distributions, identifies peak bandwidth-delay products, and records the provider service’s committed rate and restoration objective. It finds that link aggregation can increase aggregate capacity but cannot make one flow use every member. It also identifies that a nominal 1500-byte underlay cannot carry an unadjusted 1500-byte inner packet after tunnel encapsulation.

Approach and solution

The access layer is divided into purpose-specific VLANs, with inter-VLAN policy enforced by redundant Layer-3 switches. Trunks use explicit VLAN allow-lists. LACP bundles are deployed on high-volume uplinks, and traffic is monitored for hash imbalance. The inter-site provider service uses carrier Ethernet with OAM and an MPLS-based provider core. IPv6 dual stack is introduced first for managed research systems, while NAT64 is used selectively for IPv6-only client segments requiring access to legacy IPv4 services. Tunnel endpoints advertise an appropriate reduced MTU, and ICMP/ICMPv6 error messages are permitted through security controls.

Outcome

Broadcast scope and fault domains are reduced, aggregate uplink capacity improves, and service restoration becomes measurable. IPv6 applications operate without requiring a universal tunnel. The design team documents that redundancy, segmentation, and monitoring—not any single protocol—provide the operational improvement.

Learning points

1. Layer-2 expansion should be controlled through deliberate broadcast-domain design.

2. Link aggregation improves aggregate capacity, not necessarily single-flow rate.

3. Carrier services require OAM and service-level measurement.

4. Tunnel overhead must be included in MTU calculations.

5. IPv6 transition planning must include security and application behavior.

Challenges, Governance, and Best Practices

Major challenges include configuration drift, undocumented VLANs, unstable control planes, asymmetric routing, silent MTU failures, and inadequate observability. Security concerns include VLAN hopping, unauthorized trunking, MAC flooding, rogue DHCP or router advertisements, weak management access, and insecure tunnels. Controls should include authenticated management, configuration versioning, port security where appropriate, DHCP and IPv6 RA protection, segmentation based on risk, and continuous monitoring.

Best practice is to separate the data plane, control plane, and management plane; define failure domains; test convergence and protection switching; measure delay and loss rather than relying only on link utilization; and validate both IPv4 and IPv6 policy paths. Automation and AI-assisted anomaly detection can help identify unusual MAC movement, traffic shifts, or control-plane behavior, but automated remediation should be bounded by policy, auditable, and tested against false positives. Network telemetry must also be governed because flow records and endpoint metadata may contain sensitive information.

Chapter Summary

Data-link technologies provide framing, addressing, error detection, flow control, and medium access. Sliding-window protocols overcome the bandwidth-delay limitations of stop-and-wait, while ARQ selection requires probabilistic analysis of losses, retransmission cost, delay, and buffering. MAC protocols trade simplicity, efficiency, fairness, and determinism; classical ALOHA results are useful models but rely on explicit assumptions.

Ethernet switches learn source MAC addresses and use the resulting database for destination lookup. VLANs limit broadcast domains, spanning tree prevents loops, and LACP aggregates links while retaining per-flow hashing constraints. MPLS separates forwarding labels from service and routing functions, but its scalability depends on control-plane state, VPN growth, signaling, convergence, and operational complexity. Carrier Ethernet adds provider-grade service models and OAM. Tunnelling enables interoperability but introduces overhead, MTU risks, and security considerations. IPv4/IPv6 coexistence requires coordinated routing, DNS, security, monitoring, and application planning.

Key Terms

Framing: Encapsulation of a network-layer packet into a data-link-layer unit.

FCS: A frame check sequence used primarily for error detection.

CRC: Polynomial-based error-detection code.

Flow control: Regulation of sender transmission to match receiver capacity.

Sliding window: A protocol mechanism allowing multiple outstanding frames.

ARQ: Error recovery using acknowledgments, timers, and retransmissions.

Go-back-N: ARQ scheme retransmitting a missing frame and subsequent frames.

Selective repeat: ARQ scheme retransmitting only missing or damaged frames.

MAC protocol: Rule set governing access to a shared transmission medium.

Switching database: Table learned from source MAC addresses and used for destination forwarding decisions.

VLAN: Logically separated Layer-2 broadcast domain.

802.1Q trunk: Link carrying tagged frames for multiple VLANs.

LACP: Protocol for negotiating and monitoring link aggregation.

MPLS: Label-based forwarding technology supporting traffic and service separation.

FEC: Forwarding equivalence class receiving common forwarding treatment.

LSP: Label-switched path through an MPLS network.

Carrier Ethernet: Provider-grade Ethernet service with defined performance and OAM.

Tunnelling: Encapsulation of one protocol within another.

Path MTU: Largest packet that can traverse a path without fragmentation.

Dual stack: Simultaneous operation of IPv4 and IPv6.

Review Questions

Short Answer Questions

1. Why does an Ethernet switch learn source rather than destination MAC addresses?

2. Distinguish error detection from error correction.

3. What is the principal operational difference between GBN and SR ARQ?

4. Why does link aggregation not guarantee that one flow receives the sum of all member-link rates?

5. What problem does a VLAN solve, and what problem does it not automatically solve?

Long Answer Questions

1. Explain framing methods and compare their synchronization and error-resilience properties.

2. Derive the stop-and-wait utilization relationship and discuss its limitations.

3. Compare ALOHA, CSMA/CD, and CSMA/CA in terms of assumptions, collision behavior, and modern applicability.

4. Explain MPLS label operations, control-plane protocols, and scalability constraints.

5. Discuss carrier Ethernet, tunnelling, and IPv4/IPv6 transition mechanisms in a multi-site enterprise.

Analytical / Application Questions

1. A 1-Gbit/s link has an RTT of 40 ms and uses 12,000-bit frames. Estimate the minimum ideal sliding-window size needed to keep the link continuously busy, ignoring acknowledgment overhead. Explain why the practical window may need to be larger.

2. A network uses a 1500-byte path MTU, IPv4, and TCP without options. Calculate the maximum TCP data in one packet and explain how a tunnel adding 50 bytes changes the inner-packet design.

3. A provider-edge MPLS network is growing from 20 to 200 edge routers. Analyze the control-plane, route-state, and operational implications of a full-mesh design and recommend scalable alternatives.

Multiple Choice Questions with Answers

Question 1

Which address does an Ethernet switch normally learn from an incoming frame?

A. The destination MAC address

B. The source MAC address

C. The source IP address only

D. The TCP port number

Correct Answer: B. The source MAC address

Question 2

Under the idealized pure ALOHA model, the maximum throughput is approximately:

A. 0.0184

B. 0.184

C. 0.368

D. 0.500

Correct Answer: B. 0.184

Question 3

If the independent frame-error probability is 0.1 and unlimited stop-and-wait ARQ retransmissions are allowed, the expected number of transmissions per successful frame is:

A. 0.90

B. 1.00

C. 1.11

D. 1.90

Correct Answer: C. 1.11

Question 4

Which ARQ scheme retransmits only frames that are missing or damaged?

A. Stop-and-wait without sequence numbers

B. Go-back-N

C. Selective repeat

D. Pure ALOHA

Correct Answer: C. Selective repeat

Question 5

What is the principal purpose of a VLAN?

A. To increase the physical speed of every link

B. To create a logically separate Layer-2 broadcast domain

C. To encrypt all frames automatically

D. To replace IP routing entirely

Correct Answer: B. To create a logically separate Layer-2 broadcast domain

Question 6

LACP is primarily associated with:

A. IPv6 address assignment

B. Link aggregation negotiation and monitoring

C. CRC generation

D. MPLS label encryption

Correct Answer: B. Link aggregation negotiation and monitoring

Question 7

In MPLS, a transit router commonly forwards a packet by:

A. Swapping an incoming label for an outgoing label

B. Recomputing the TCP checksum only

C. Converting IPv4 directly into Ethernet broadcast

D. Discarding every packet with a label

Correct Answer: A. Swapping an incoming label for an outgoing label

Question 8

For a 1500-byte MTU and a 40-byte IPv4-plus-TCP header total, the maximum TCP data payload is:

A. 1400 bytes

B. 1450 bytes

C. 1460 bytes

D. 1480 bytes

Correct Answer: C. 1460 bytes

Question 9

Which statement about IPv6 transit fragmentation is correct?

A. Every IPv6 router fragments packets automatically

B. IPv6 has no mechanism for fragmentation at all

C. The source performs fragmentation when required, using an extension header

D. Only Ethernet switches perform IPv6 fragmentation

Correct Answer: C. The source performs fragmentation when required, using an extension header

Question 10

A major scalability concern in a full mesh of n edge routers is that the number of pairwise sessions is:

A. n

B. 2n

C. n squared

D. n(n−1)/2

Correct Answer: D. n(n−1)/2