#3 Network Core and Packet Switching Vocabulary

Course Overview and Scope

• Core Focus: Comprehensive examination of the structure and internal mechanics of the Internet, specifically emphasizing the network core and packet-switching paradigms.

• Curricular Mapping: Aligned with textbook Chapters 1.2 – 1.3.

Fundamental Network-Core Functions: Routing vs. Forwarding

• Two Primary Core Functions: The network core relies on two distinct but interrelated operations to move data end-to-end across interconnected nodes: forwarding and routing.

• Forwarding (Switching):

  • Scope: Local action performed individually by each router.

  • Definition: The physical transfer of an arriving packet from a router's input link interface to the appropriate output link interface.

  • Execution Protocol: An arriving packet contains a header with specific control information. The router reads the destination address in the packet's header (e.g., header value $0111$) and uses it to index into its local forwarding table to determine the designated output link (e.g., output link $3$).

  • Local Forwarding Table Mapping Example:

  • Header value $0100 \rightarrow$ Output link $3$

  • Header value $0101 \rightarrow$ Output link $2$

  • Header value $0111 \rightarrow$ Output link $3$

  • Header value $1001 \rightarrow$ Output link $1$

  • Analogy: Making a local turn or taking a specific exit ramp at a single highway intersection (e.g., local navigation around Sacramento sub-regions like North Natomas, Arden-Arcade, and Rosemont).

• Routing:

  • Scope: Global action spanning the entire network infrastructure.

  • Definition: The process of determining the complete, end-to-end path that packets travel from a source host to a destination host.

  • Execution Protocol: Driven by routing algorithms that dynamically compute optimal paths across interconnected networks and write these routing decisions directly into each router's local forwarding table.

  • Analogy: Planning a long-distance road trip route across an entire nation

Packet-Switching Architecture and Mechanics

• Store-and-Forward Transmission:

  • Transmission Rule: Packet-switched networks employ store-and-forward transmission, meaning an intermediate router must receive an entire packet before it can begin transmitting the first bit of that packet onto the outgoing link.

  • Packet Transmission Delay Formula:     $D_{\text{trans}} = \frac{L}{R}$     where $L$ represents the length of the packet in bits, and $R$ represents the link transmission rate (bandwidth) in bits per second ($\text{bps}$).

  • Concrete One-Hop Calculation Example:

  • Packet length $L = 10\,\text{Kbits} = 10,000\,\text{bits}$

  • Link transmission rate $R = 100\,\text{Mbps} = 100,000,000\,\text{bps}$

  • One-hop transmission delay calculation:       $D_{\text{trans}} = \frac{10\,\text{Kbits}}{100\,\text{Mbps}} = \frac{10,000\,\text{bits}}{100,000,000\,\text{bps}} = 0.0001\,\text{sec} = 0.1\,\text{msec}$

• Queueing Delay and Packet Loss:

  • Queueing Cause: Queueing occurs when the rate at which data arrives at a router link exceeds the link's transmission capacity over a given period.

  • Router Queueing Dynamics:

  • Consider sources $S_1$ and $S_2$ connected to router $R_1$ over $100\,\text{Mb/s}$ links, with $R_1$ forwarding traffic to $R_2$ over a $1.5\,\text{Mb/s}$ bottleneck link.

  • If the incoming workload rate exceeds $1.5\,\text{Mb/s}$, packets must wait in an output queue (buffer) at router $R_1$ before being pushed onto the link.

  • Packet Loss Condition: Router memory buffers are finite. If packets arrive at a rate higher than the link transmission capability for an extended period, the memory buffer fills completely, and newly arriving packets are dropped (lost).

    • Packets are dropped based on which packet was there the longest and if the buffer full.

Circuit Switching vs. Packet Switching

• Circuit Switching Principles:

  • Resource Reservation: End-to-end network resources (bandwidth, link capacity) are explicitly allocated and reserved for the entire duration of a communication session ("call") between source and destination.

  • Multiplexing Structure: Links are divided into distinct circuits (e.g., $4$ circuits per link). A call is assigned a specific circuit segment along every link in its path (e.g., using the $2\text{nd}$ circuit in a top link and the $1\text{st}$ circuit in a right link).

  • Characteristics:

  • Fully dedicated resources with zero resource sharing.

  • Guaranteed end-to-end performance stability.

  • Wasteful during silent periods: Circuit segments remain idle if no data is currently being transmitted by the active call.

  • Traditional foundation of legacy telephone networks.

• Quantitative Comparison Example:

  • System Parameters:

  • Shared link rate: $1\,\text{Gbps} = 1,000\,\text{Mbps}$

  • Per-user activity demand: Each user generates $100\,\text{Mb/s}$ when active.

  • User activity factor: Each user is active only $10\%$ of the time ($p = 0.1$).

  • Circuit Switching Capacity:     $\text{Maximum Users} = \frac{1\,\text{Gbps}}{100\,\text{Mb/s}} = \frac{1,000\,\text{Mbps}}{100\,\text{Mbps}} = 10\,\text{users}$     Circuit switching can support a maximum of exactly $N = 10$ users, regardless of how infrequently individual users transmit data.

  • Packet Switching Capacity:

  • With $N = 35$ total users utilizing packet switching on the same link:

  • The probability that more than $10$ users are active simultaneously is extremely low:       P(\text{Active Users} > 10) < 0.0004

  • Packet switching successfully supports $35$ users (more than $3\times$ circuit switching capacity) with negligible probability of congestion.

• Trade-Off Analysis:

  • Packet Switching Advantages:

    • Superior efficiency for bursty, unpredictable data traffic through statistical multiplexing.

    • Significantly simpler call setup with no mandatory prior resource reservation step.

  • Packet Switching Disadvantages:

    • Susceptible to network congestion leading to variable queueing delays and packet drops due to buffer overflow.

    • Requires explicit transport protocols to handle reliable data transfer and congestion control.

    • Achieving circuit-like performance guarantees over packet-switched networks is possible but complex (addressed across three dedicated TCP lectures).

Internet Structure: A Network of Networks

• Access Networks and Interconnection Needs:

  • Individual end-hosts, home networks, mobile networks, enterprise networks, and datacenter networks connect to the Internet via access Internet Service Providers (ISPs).

  • To enable any arbitrary pair of hosts worldwide to communicate, access ISPs must be fully interconnected.

  • Economic factors, technological limitations, and national policies drive the structural evolution of this complex network of networks.

• Scalability and Evolutionary Topologies:

  • Mesh Topology Limit: Directly interconnecting every access ISP to every other access ISP requires $O(N^2)$ individual connections, which is completely unscalable for millions of access ISPs.

  • Global Transit ISP Option: Connecting every access ISP to a single global transit ISP creates a functional system, but a single global provider inevitably introduces competition.

  • Multi-ISP Coexistence: When multiple global ISPs exist (e.g., ISP A, ISP B, ISP C), they must interconnect with each other:

  • Peering Links: Direct interconnects between competing ISPs.

  • Internet Exchange Points (IXPs): Specialized third-party physical points where multiple ISPs gather to exchange traffic directly.

  • Regional ISPs: Arise as intermediate entities to aggregate local access ISPs and connect them to larger global/tier-1 networks.

• Core Structural Hierarchy:

  • Tier-1 Commercial ISPs: Operating at the center of the Internet, these networks provide global and national coverage (e.g., Level 3, Sprint, AT&T).

  • Utilize Points-of-Presence (PoPs): Interconnection sites that join Sprint/Tier-1 networks with customer networks, peering networks, and other internal PoPs.

  • Content Provider Networks: Private corporate backbones operated by tech giants (e.g., Google, Akamai, Amazon).

    • These run packets parallel within their own ISP first then push outwards to the destination.

  • Connect proprietary data centers directly to the broader Internet, bypassing traditional regional and Tier-1 ISPs to optimize latency and control routing costs.

Infrastructure Case Studies

• Tier-1 ISP Architecture (Sprint):

  • Consists of a global backbone connecting internal PoPs via high-speed links.

  • Links directly to customer networks, competing peering networks, and external infrastructure.

• Global Content Provider Architecture (Google Network):

  • Features over $100$ Points of Presence (PoPs) globally.

  • Deploys extensive private network fiber and localized Google Global Cache (GGC) edge nodes to serve end-users directly.

  • Operates dedicated international subsea fiber optic cable systems:

  • Unity: Deployed in $2010$, connecting the United States and Japan.

  • SJC: Deployed in $2013$, connecting Japan, Hong Kong, and Singapore.

  • FASTER: Deployed in $2016$, connecting the United States, Japan, and Taiwan.

  • Monet: Deployed in $2017$, connecting the United States and Brazil.