Course: COMP347 Computer Networks (Revision 10) | Textbook: Kurose & Ross, Computer Networking: A Top-Down Approach (9th ed.)
Upon completion of this tutorial, students should be able to:
The network layer is the third layer of the Internet protocol stack, responsible for moving data segments from a source host to a destination host across multiple networks. It sits between the transport layer (which provides end‑to‑end communication services) and the link layer (which handles communication over a single link).
This tutorial introduces the foundational concepts of the network layer, with particular emphasis on the data plane—the per‑router function that determines how datagrams arriving on a router's input port are forwarded to the appropriate output port. Understanding the data plane is essential for grasping how packets are actually moved through the Internet's infrastructure.
The network layer is traditionally divided into two functional planes:
| Plane | Function | Scope | Implementation |
|---|---|---|---|
| Data Plane | Per‑router forwarding decisions | Local, per‑router | Hardware (typically) |
| Control Plane | Network‑wide routing decisions | Global, network‑wide | Software |
This separation is fundamental to modern network design and is a recurring theme throughout this unit. We will explore the architecture, protocols, and mechanisms that enable the data plane to operate at wire speed while the control plane manages the complexity of global routing.
The network layer's primary responsibility is to transport data segments from a sending host to a receiving host. Unlike the transport layer, which provides communication between processes, the network layer provides communication between hosts.
Key functions include:
Every Internet device—including hosts and routers—implements the network layer. The network layer provides a uniform interface (the IP protocol) that hides the heterogeneity of underlying link‑layer technologies.
In addition to the core functions, the network layer may also support:
Two core network‑layer functions are often confused but are conceptually distinct:
Forwarding refers to the local action of moving a packet from a router's input link to the appropriate output link. When a packet arrives at a router's input port, the router must examine the packet's header, consult its forwarding table, and determine which output port to use.
Forwarding is:
Routing refers to the network‑wide process of determining the end‑to‑end path that packets should take from source to destination. Routing algorithms compute the paths that forwarding tables will use.
Routing is:
The data plane is the local, per‑router function that determines how a datagram arriving on a router's input port is forwarded to the router's output port.
Characteristics:
The control plane is the network‑wide logic that determines how datagrams are routed among routers along the end‑to‑end path from source to destination.
Two approaches to the control plane exist:
The network layer can provide various service models, depending on the network architecture. The service model defines the characteristics of the "channel" that transports datagrams from sender to receiver.
| Service Attribute | Description |
|---|---|
| Guaranteed delivery | The network ensures every datagram reaches its destination |
| Guaranteed delivery with bounded delay | Datagrams arrive within a specified time |
| In‑order delivery | Datagrams arrive in the order they were sent |
| Guaranteed minimum bandwidth | A minimum throughput is assured for a flow |
| Restrictions on inter‑packet spacing | The network preserves timing between packets |
| Security guarantees | Confidentiality, integrity, or authentication services |
The Internet's network layer provides a best‑effort service model.
Under best‑effort service:
This is a conscious design choice that simplifies the network layer and allows the Internet to scale. Higher‑layer protocols (particularly TCP) provide reliability and ordering when needed.
| Network Architecture | Service Model | Bandwidth Guarantee | Loss Guarantee | Order Guarantee | Timing Guarantee |
|---|---|---|---|---|---|
| Internet | Best effort | None | No | No | No |
| ATM CBR | Constant bit rate | Yes | Yes | Yes | Yes |
| ATM VBR | Variable bit rate | Yes | Yes | Yes | Yes |
| Frame Relay | Frame‑based | May be guaranteed | No (discard eligible) | No | No |
| MPLS | Label‑switched | Possible via TE | No (best effort by default) | No (but often preserved) | Possible via TE |
In a datagram network, each packet is forwarded independently using the destination address in the packet header.
Characteristics:
Forwarding Process:
In a virtual‑circuit network, a connection (virtual circuit) is established before data transfer begins.
Characteristics:
Three Phases:
| Aspect | Datagram Networks | Virtual‑Circuit Networks |
|---|---|---|
| Connection setup | Not required | Required |
| Per‑flow state | None in routers | Maintained in routers |
| Addressing | Destination address in each packet | VC identifier in each packet |
| Path | May vary per packet | Fixed for entire connection |
| Router complexity | Simple forwarding | More complex (VC table, label swapping) |
| QoS support | Difficult | Easier (resources reserved at setup) |
| Internet use | Yes (IP) | Limited (MPLS, VPNs) |
The Internet's network layer is built around the Internet Protocol (IP), which provides the datagram (connectionless) service model.
Key components:
The Internet network layer is designed to be:
The Internet's network layer has been shaped by several key design principles:
Functions that can be implemented at the endpoints should not be implemented in the network core. This principle argues that the network should provide a simple, best‑effort service, leaving reliability and other complex functions to the end hosts. This reduces network complexity and allows applications to choose the level of reliability they need.
State required for communication should be stored at the endpoints rather than in the network core. If a network element fails, only ongoing connections that depend on that element are affected. This improves robustness and simplifies recovery.
The network layer separates concerns through a well‑defined interface, allowing independent evolution of the network layer and higher layers (transport, application).
The network layer is designed to run over diverse link‑layer technologies (Ethernet, Wi‑Fi, optical, etc.) and to support a wide range of applications. This "hourglass" architecture has been key to the Internet's success.
MPLS is a network‑layer technology that operates between the network and link layers. It uses labels to forward packets along predetermined paths (Label Switched Paths, LSPs). MPLS provides a hybrid approach: it is connection‑oriented at the network layer but can carry IP datagrams. It enables traffic engineering, VPNs, and QoS.
SDN decouples the control plane from the data plane, placing the control logic in a centralized controller. The data plane devices become simple forwarders that execute rules received from the controller. This enables programmable networks, faster innovation, and simplified management.
Emerging technologies allow the data plane to be reprogrammed dynamically, enabling new packet‑processing pipelines without hardware changes. This extends the flexibility of the network layer beyond traditional IP forwarding.
Q1. What is the primary difference between forwarding and routing?
C) Forwarding is the local action of moving packets from input to output ports on a single router; routing is the network‑wide process of determining the path.
Q2. Which of the following is NOT a characteristic of the Internet's best‑effort service model?
C) The Internet's best‑effort service does not guarantee minimum bandwidth.
Q3. The data plane operates at which timescale?
C) The data plane processes packets individually at extremely high speed, typically nanoseconds to microseconds.
Q4. In a virtual‑circuit network, what is the correct sequence of phases?
C) Virtual‑circuit networks require setup, then data transfer, then teardown.
Q5. Which plane is responsible for determining how a datagram is forwarded from a router's input port to its output port?
B) The data plane handles the per‑router forwarding decision.
Q6. Which of the following is an example of a connection‑oriented network‑layer technology?
B) ATM (Asynchronous Transfer Mode) uses virtual circuits.
Q7. In an SDN architecture, where does the control plane typically run?
B) In SDN, the control plane is centralized on one or more controllers.
Q8. Which design principle argues that reliability should be implemented at the endpoints rather than in the network core?
B) The end‑to‑end argument suggests that network core should be simple and reliability handled by endpoints.
Q9. Which of the following best describes a datagram network?
C) In a datagram network, each packet is forwarded independently based on destination address.
Q10. What is the primary purpose of the routing processor in a router?
B) The routing processor handles control‑plane functions, including routing protocols and forwarding table computation.
Q11. Define the term "forwarding table" and explain how it is used in the data plane.
A forwarding table is a data structure maintained by a router that maps destination addresses (or address prefixes) to output ports. When a packet arrives, the router consults the forwarding table to determine the appropriate output port. It is used in the data plane for per‑packet forwarding decisions.
Q12. Why does the Internet use a connectionless (datagram) network layer rather than a connection‑oriented one? Give two reasons.
1. Scalability: No per‑flow state is needed in routers, allowing them to handle huge numbers of flows.
2. Robustness: If a router fails, packets can be rerouted without needing to re‑establish connections, making the network more resilient.
Q13. Describe the end‑to‑end argument and give an example of how it applies to the Internet.
The end‑to‑end argument states that functions that can be implemented correctly at the endpoints should not be implemented in the network core. For example, reliable data transfer is implemented by TCP at the transport layer (end hosts) rather than by the IP network layer. This keeps the network simple and allows applications to choose their own reliability mechanisms.
Q14. What is the difference between the data plane and the control plane in terms of implementation technology?
The data plane is typically implemented in dedicated hardware (ASICs, TCAM, network processors) to achieve high‑speed packet processing. The control plane is implemented in software running on general‑purpose CPUs, as it involves complex routing computations and network‑wide coordination.
Q15. Explain why the Internet's network layer is often described as an "hourglass" architecture.
The "hourglass" architecture refers to the narrow waist of the Internet stack at the IP layer. IP provides a common interface that can run over many different link‑layer technologies (below) and support many transport/application protocols (above). This design allows for heterogeneity and innovation at both the lower and upper layers while maintaining a stable core.
Q16. Consider a router that receives a packet with destination address 192.168.5.10. Its forwarding table has entries: 192.168.0.0/16 → Port 1, 192.168.5.0/24 → Port 2, and 0.0.0.0/0 → Port 3. Which output port is used, and why?
The router will use Port 2 because it uses longest‑prefix matching. The address matches 192.168.0.0/16 (16 bits) and 192.168.5.0/24 (24 bits), and the longer prefix (24) is preferred.
Q17. A network operator wants to provide guaranteed bandwidth for VoIP traffic. Which network‑layer service model (from the list: best‑effort, CBR, VBR) would be most suitable, and why?
CBR (Constant Bit Rate) would be most suitable because it provides a fixed bandwidth and low delay, which are essential for real‑time VoIP. Best‑effort does not guarantee bandwidth, and VBR may not provide the constant rate needed.
Q18. In an SDN environment, a network administrator wants to implement a new forwarding policy that requires inspecting packet payloads. How would this be achieved in the data plane?
With programmable data planes (e.g., using P4), the administrator can modify the packet processing pipeline to inspect payload fields. The SDN controller would install new flow rules in the switches that match on the new fields and specify actions, enabling flexible forwarding policies without hardware changes.
Q19. Compare the reliability of a datagram network versus a virtual‑circuit network in the event of a link failure.
In a datagram network, if a link fails, routers can dynamically reroute subsequent packets around the failure because each packet is forwarded independently. Existing flows may experience transient loss but can recover quickly. In a virtual‑circuit network, the failure disrupts the established path; the connection must be torn down and re‑established, causing a longer service interruption. Thus, datagram networks are generally more robust to failures.
Q20. A router has a forwarding table that uses longest‑prefix matching. Explain why this is necessary in the context of CIDR addressing.
CIDR (Classless Inter‑Domain Routing) allows variable‑length prefixes, so multiple entries may match a given destination address. Longest‑prefix matching ensures that the most specific route (i.e., the one with the longest prefix) is selected, which allows for route aggregation and efficient address allocation. Without longest‑prefix matching, routers could not correctly choose the most specific route.
Q21. Which of the following is a characteristic of a virtual‑circuit network but not a datagram network?
B) Virtual‑circuit networks maintain per‑flow state (VC tables) in routers, whereas datagram networks do not.
Q22. What is the role of the forwarding table in the data plane?
B) The forwarding table is used to map destination addresses (or prefixes) to output ports for packet forwarding.
A router receives a packet with destination IP address 10.20.30.40. The forwarding table contains the following entries:
Which port will be used and why?
The longest matching prefix is 10.20.30.32/27 (27 bits). The destination 10.20.30.40 falls in the range 10.20.30.32 – 10.20.30.63, so it matches. Therefore, the packet is forwarded to Port C.
Explain the difference between input buffering and output buffering in a router. Which one suffers from Head‑of‑Line (HOL) blocking and why?
Input buffering: Packets are queued at input ports when the switching fabric is busy. It suffers from HOL blocking because a packet at the front of an input queue destined for a busy output can block all packets behind it, even if they are destined for available outputs.
Output buffering: Packets are queued at output ports when the output link is busy. It does not suffer from HOL blocking because each output queue is independent; however, it may require more buffer memory.
A network has a mix of real‑time voice traffic and elastic data traffic. Discuss which network‑layer service model (best‑effort, CBR, or VBR) would be appropriate for each type of traffic, and explain the trade‑offs.
Voice: CBR is appropriate because voice requires a constant bit rate, low delay, and low jitter. Guaranteed bandwidth and timing are essential.
Data: Best‑effort is usually sufficient because data applications are tolerant of delay and can adapt to congestion via TCP. Elastic traffic benefits from the simplicity and scalability of best‑effort.
Trade‑off: CBR requires resource reservation and more complex network management, while best‑effort is simpler but cannot guarantee quality for real‑time services.
Suppose a router uses a crossbar switch fabric with 8 input ports and 8 output ports. The switching speed is 50 ns per packet. If each packet is 1000 bytes, what is the maximum aggregate throughput (in Gbps) if only one packet can be switched at a time? What if the crossbar can switch up to 4 packets simultaneously?
One packet at a time:
Switching time = 50 ns = 5×10⁻⁸ s.
Packets per second = 1 / (5×10⁻⁸) = 20,000,000 pps.
Throughput = 20,000,000 × 1000 bytes × 8 bits/byte = 160,000,000,000 bits/s = 160 Gbps.
Up to 4 packets simultaneously:
Aggregate throughput = 4 × 160 Gbps = 640 Gbps.
Describe the path of an IP datagram from a web server to a client, listing all network‑layer operations at each router and host along the way. Include encapsulation, forwarding, and decapsulation.
At the server:
1. The transport layer (TCP) sends a segment to the network layer.
2. The network layer encapsulates the segment into an IP datagram, adding the IP header (source = server IP, destination = client IP).
3. The datagram is passed to the link layer for transmission on the outgoing link.
At each intermediate router:
1. The link layer receives the frame, extracts the IP datagram.
2. The input port looks up the destination IP in the forwarding table (using longest‑prefix matching).
3. The datagram is switched to the appropriate output port (via switching fabric).
4. The output port decrements TTL, recomputes checksum, and passes to link layer for transmission on the next link.
At the client:
1. The link layer receives the frame and extracts the IP datagram.
2. The network layer verifies the header checksum and delivers the payload (TCP segment) to the transport layer.
Compare the per‑packet processing overhead in a datagram network versus a virtual‑circuit network. Which requires more processing per packet and why?
In a datagram network, each packet requires a full forwarding table lookup based on the destination address (longest‑prefix matching). This lookup can be computationally intensive. In a virtual‑circuit network, the forwarding is based on a simple VC label lookup, which is faster (like a table index). However, the setup phase in VC networks is more complex. Per‑packet, VC forwarding is typically faster because label lookup is simpler than address lookup. But the overall system complexity is higher due to setup and teardown.
Explain why the separation of data plane and control plane is important for network scalability and evolution. Give two examples of how this separation enables innovation.
Separation allows the data plane to be optimized for speed while the control plane can be updated independently. This enables:
A router's forwarding table has the following entries (prefix: output port):
172.16.0.0/12 → P1
172.16.128.0/17 → P2
172.16.128.0/18 → P3
0.0.0.0/0 → P4
For each of the following destination addresses, determine the output port:
a) 172.16.129.10
b) 172.16.64.5
c) 172.16.200.3
a) 172.16.129.10: matches 172.16.0.0/12, 172.16.128.0/17, and 172.16.128.0/18. Longest match is /18 → P3.
b) 172.16.64.5: matches 172.16.0.0/12 only → P1.
c) 172.16.200.3: matches 172.16.0.0/12 and 172.16.128.0/17 (but not /18). Longest is /17 → P2.
Describe the concept of "fate sharing" and give an example of how it applies to the Internet's network layer.
Fate sharing means that the state required for a communication session is stored at the endpoints, not in the network. If a network element (router) fails, only those sessions that are currently using that element are affected, and they can recover using endpoint mechanisms. For example, TCP connections store state at the hosts; if a router in the middle fails, the hosts can retransmit and reroute around the failure without requiring the network to maintain per‑flow state.
Consider a scenario where a network operator wants to implement a new policy that prioritizes certain types of traffic (e.g., VoIP) over others. How would this be implemented in a traditional network versus an SDN‑enabled network?
Traditional: The operator would configure DiffServ (Differentiated Services) on each router, marking packets with DSCP values and configuring per‑hop behaviors. This requires manual configuration on every router and is complex to manage.
SDN: The operator would program a centralized controller to detect VoIP flows (using deep packet inspection or flow classification) and install flow rules in switches that give higher priority to those flows. The controller can dynamically adjust policies and push rules to all switches centrally, simplifying management and enabling rapid changes.
Research the network‑layer service models of ATM (Asynchronous Transfer Mode) and Frame Relay. Compare them with the Internet's best‑effort service. Write a report (approx. 500 words) that includes a table summarizing the key service attributes (bandwidth, delay, loss, ordering) for each.
ATM Service Categories:
- CBR: Constant bit rate, low delay, low jitter, suitable for voice/video.
- VBR: Variable bit rate, statistical multiplexing, suitable for compressed video.
- ABR: Available bit rate, minimum cell rate guarantee if available.
- UBR: Unspecified bit rate, best‑effort similar to IP.
Frame Relay: Provides a connection‑oriented service with Committed Information Rate (CIR) and Excess Information Rate (EIR). Frames exceeding CIR may be marked as discard eligible (DE) and dropped in congestion. No order or timing guarantees.
Comparison Table:
| Attribute | Internet Best‑Effort | ATM CBR | ATM VBR | Frame Relay |
|-----------|----------------------|---------|---------|-------------|
| Bandwidth Guarantee | None | Yes (CBR) | Yes (VBR) | CIR only |
| Delay Guarantee | None | Yes | Yes (bounded) | No |
| Loss Guarantee | No | Very low | Low | No (DE may be dropped) |
| Ordering | No | Yes | Yes | No (but often preserved) |
Explain the concept of "longest‑prefix matching" and why it is essential for CIDR. Give an example illustrating how a router chooses the correct output port when multiple prefixes match a destination address.
Longest‑prefix matching is the process by which a router selects the forwarding table entry with the longest prefix that matches the destination address. CIDR allows arbitrary prefix lengths, so multiple entries may match. The router chooses the most specific (longest) match to ensure packets are sent to the correct next hop.
Example: Destination 192.168.1.5. Table entries: 192.168.0.0/16 → P1, 192.168.1.0/24 → P2, 192.168.1.0/28 → P3. The /28 match is longest, so P3 is used.
Design a small network with three routers (R1, R2, R3) connected in a chain. Write the forwarding tables for each router, assuming the following subnets: Subnet A (10.0.1.0/24) connected to R1, Subnet B (10.0.2.0/24) connected to R2, Subnet C (10.0.3.0/24) connected to R3. The links between routers are point‑to‑point with addresses 10.0.4.0/30 (R1‑R2) and 10.0.5.0/30 (R2‑R3). Include default routes if appropriate.
R1 Forwarding Table:
- 10.0.1.0/24 → directly connected (Port A)
- 10.0.2.0/24 → next hop 10.0.4.2 (via link to R2)
- 10.0.3.0/24 → next hop 10.0.4.2 (via R2)
- 0.0.0.0/0 → next hop 10.0.4.2 (default)
R2 Forwarding Table:
- 10.0.2.0/24 → directly connected (Port B)
- 10.0.1.0/24 → next hop 10.0.4.1 (via R1)
- 10.0.3.0/24 → next hop 10.0.5.2 (via R3)
- 0.0.0.0/0 → next hop 10.0.4.1 (or 10.0.5.2 depending on policy)
R3 Forwarding Table:
- 10.0.3.0/24 → directly connected (Port C)
- 10.0.1.0/24 → next hop 10.0.5.1 (via R2)
- 10.0.2.0/24 → next hop 10.0.5.1 (via R2)
- 0.0.0.0/0 → next hop 10.0.5.1
Compare and contrast the following pairs of terms in the context of the network layer:
a) Forwarding and routing
b) Data plane and control plane
c) Datagram networks and virtual‑circuit networks
a) Forwarding vs. Routing: Forwarding is the local per‑router action of moving packets from input to output ports based on forwarding tables. Routing is the network‑wide process of determining the paths that packets take, involving routing protocols and algorithms. Forwarding is fast (hardware), routing is slower (software).
b) Data plane vs. Control plane: Data plane handles per‑packet forwarding decisions and operates at very high speed. Control plane handles network‑wide routing decisions, management, and configuration. The data plane uses forwarding tables; the control plane computes them.
c) Datagram vs. Virtual‑circuit: Datagram networks are connectionless, each packet forwarded independently using destination addresses; no per‑flow state. Virtual‑circuit networks require a setup phase, maintain per‑flow state, and forward using a VC label; they can support QoS but are more complex.
Describe how the network layer supports the transport layer's reliable data transfer. In particular, explain the interaction between the network layer (IP) and the transport layer (TCP) in terms of segmentation, encapsulation, and error detection.
TCP segments are passed to the network layer (IP) for transmission. The network layer encapsulates each segment into an IP datagram, adding a header with source/destination IP addresses, TTL, etc. The network layer does not guarantee delivery, order, or integrity (only header checksum). TCP provides reliability by using sequence numbers, acknowledgments, retransmissions, and checksums at the transport layer. The network layer's best‑effort service allows TCP to implement its own reliability mechanisms without relying on the network. The network layer also fragments large datagrams if needed, and TCP can adjust its segment size based on the path MTU (discovered through ICMP).
Research the concept of "active queue management" (AQM) and its relationship to the data plane. How does AQM affect forwarding decisions and congestion control?
Active Queue Management (AQM) is a mechanism that proactively drops packets or marks them (e.g., ECN) before the queue becomes full, to signal congestion to TCP senders. In the data plane, AQM is implemented in the output port's buffering and scheduling logic. Instead of tail drop (dropping only when full), AQM drops packets probabilistically based on queue length (RED) or other metrics. This helps to avoid global synchronization of TCP flows and maintains lower average queue lengths. AQM is a data‑plane feature that affects forwarding because it determines which packets are dropped, influencing overall throughput and fairness.
Explain the role of the Internet Control Message Protocol (ICMP) in the network layer. Give examples of ICMP messages that are used for error reporting and diagnostics, and relate them to the data plane.
ICMP is used by the network layer for error reporting and diagnostics. ICMP messages are encapsulated in IP datagrams and are generated by routers or hosts in response to problems. Examples: Destination Unreachable (e.g., network unreachable, host unreachable, port unreachable); Time Exceeded (TTL expired, used by traceroute); Echo Request/Reply (used by ping). These messages help the data plane by informing hosts about forwarding problems; they are not directly part of forwarding but are essential for network troubleshooting and path discovery.
Discuss the scalability challenges of the Internet's network layer and how technologies like CIDR, NAT, and IPv6 address these challenges.
Scalability challenges: The exponential growth of the Internet has led to routing table growth and exhaustion of IPv4 addresses.
CIDR: Allows address aggregation, reducing the size of routing tables by grouping contiguous addresses.
NAT: Conserves public IPv4 addresses by allowing private networks to use private addresses and share a single public address.
IPv6: Provides a vastly larger address space, eliminating the need for NAT and enabling more efficient routing. IPv6 also simplifies header processing, which can improve data plane performance.
Explain the concept of "software‑defined networking" (SDN) and how it changes the relationship between the data plane and control plane. What are the benefits and potential drawbacks?
SDN decouples the control plane from the data plane by centralizing control logic in a software controller that communicates with the forwarding devices (switches/routers) using a standard protocol like OpenFlow. The data plane devices become simple forwarders that execute rules installed by the controller. Benefits include: programmability, easier network management, rapid innovation, and global visibility. Drawbacks include: potential single point of failure (controller), scalability of the controller, security concerns, and the need for a reliable control‑channel.
Write a short essay (approx. 300 words) on the evolution of the network layer from IPv4 to IPv6, focusing on the changes that affect the data plane (e.g., header simplification, addressing, and fragmentation).
IPv6 was developed to address IPv4 address exhaustion and to simplify the network layer. Key data‑plane changes include:
- Larger address space: 128‑bit addresses allow for vast scalability and eliminate NAT.
- Simplified header: Fixed 40‑byte header with fewer fields, easier to process in hardware; removed checksum, fragmentation fields moved to extension headers.
- Fragmentation: In IPv6, only source hosts fragment; routers do not fragment, simplifying forwarding.
- Extension headers: Allow for optional features (e.g., routing, fragmentation) without increasing the base header size.
These changes improve data plane efficiency and enable faster forwarding, as routers have less header processing to perform. Transition mechanisms (dual‑stack, tunneling) allow gradual deployment.
This expanded tutorial has provided a comprehensive introduction to the network layer and the data plane, covering both foundational concepts and advanced topics. Key takeaways include:
Understanding these concepts is crucial for the subsequent tutorials in this unit, which will delve deeper into router internals, forwarding algorithms, IP addressing, fragmentation, and more.