At the heart of circuit switching lies a contractual promise: once a circuit is established, the network guarantees that specific resources—bandwidth, switch capacity, transmission facilities—remain available for the duration of the session. This is resource reservation, the mechanism that transforms a network of shared infrastructure into what appears to be a private, dedicated connection.

Resource reservation is what enables the guaranteed Quality of Service that distinguishes circuit switching from best-effort packet networks. It's also the source of circuit switching's efficiency limitations. Understanding resource reservation deeply reveals both the power and the constraints of this architectural approach.

In this page, we examine exactly what resources are reserved, how reservation decisions are made, the granularity at which resources are allocated, and how reservation concepts have evolved and persisted into modern networking.

Resource reservation in circuit switching encompasses multiple categories of network resources. Each category must be available and allocated for a circuit to function correctly.

1. Transmission Bandwidth:

The most fundamental reserved resource. For each hop in the circuit path, sufficient transmission capacity must be allocated:

• In analog systems: frequency band on the carrier
• In TDM digital systems: time slots in the frame
• In WDM optical systems: wavelength or fraction thereof

2. Switch Port Capacity:

Each switch in the path allocates:

• Input port connection for incoming signal
• Output port connection for outgoing signal
• Internal cross-connect (fabric) resources to link them

3. Processing Resources:

During setup and teardown:

• Call processing capacity (signaling message handling)
• Routing table lookups
• State maintenance (call records, billing data)

Contrast with packet switching:

In packet-switched networks, resources are not reserved:

• Bandwidth is shared among all flows on a link
• Switch/router capacity is allocated per-packet based on arrivals
• Buffering absorbs temporary congestion but introduces delay
• No advance commitment means no guaranteed capacity

Circuit switching's reservation model trades statistical efficiency for predictable performance. The resources allocated to an idle circuit cannot serve other traffic—but the circuit owner enjoys guaranteed, consistent service.

In digital circuit-switched networks, Time Division Multiplexing (TDM) is the primary mechanism for creating multiple logical circuits on a single physical link. Understanding time slot assignment is essential for comprehending how resources are actually reserved.

TDM fundamentals:

TDM divides a transmission link's capacity into recurring frames, each containing multiple time slots. Each slot carries data for one circuit during each frame period:

T1 (DS-1) structure:

• Frame rate: 8,000 frames/second (125 μs per frame)
• Slots per frame: 24
• Bits per slot: 8 (64 kbps per slot)
• Total capacity: 24 × 64 = 1.536 Mbps + 8 kbps framing = 1.544 Mbps

E1 structure:

• Frame rate: 8,000 frames/second (125 μs per frame)
• Slots per frame: 32 (30 for traffic, 2 for signaling/sync)
• Bits per slot: 8 (64 kbps per slot)
• Total capacity: 32 × 64 = 2.048 Mbps

Time slot assignment process:

1. Slot selection at originating switch:

When a call is setup:

• Switch examines available slots on the outgoing trunk
• Selects an available slot (sequential, random, or load-balanced)
• Marks slot as 'in use' in trunk slot map
• Records slot number in call state

2. Slot interchange at intermediate switches:

Here's a critical complexity: the incoming slot number may differ from the available outgoing slot. Switches perform Time Slot Interchange (TSI):

Incoming: Slot 5 on Trunk A
Needed:   Slot 12 on Trunk B (slot 5 busy on B)

Switch performs:
1. Read sample from Slot 5, Trunk A
2. Buffer the sample
3. Write sample to Slot 12, Trunk B

This happens in real-time, every 125 μs, maintaining synchronous timing.

3. End-to-end slot chain:

A complete circuit might use different slot numbers on each link:

The TSI at each switch maps between these slots seamlessly.

The granularity of resource reservation determines how finely capacity can be allocated. Finer granularity improves efficiency but increases complexity. Different technologies offer different granularity options.

Historical progression:

The granularity problem:

If your application needs 100 Mbps but the smallest allocatable unit is 155 Mbps (STS-3), you waste 55 Mbps—35% overhead.

Conversely, if you need 200 Mbps, you must allocate STS-3c (155 Mbps) + STS-1 (52 Mbps) = 207 Mbps, or step up to STS-12 (622 Mbps), wasting even more.

Solutions to granularity constraints:

1. Virtual Concatenation (VCAT):

SONET Virtual Concatenation allows combining non-contiguous smaller containers:

• Need 100 Mbps? Use 2 × STS-1 (104 Mbps)
• The containers need not be adjacent in the frame
• Provides much finer granularity without new hardware

2. Link Capacity Adjustment Scheme (LCAS):

Builds on VCAT to allow dynamic adjustment:

• Increase allocation by adding containers
• Decrease by removing them
• Hitless (no service disruption during changes)
• Approaches packet-like flexibility while maintaining circuit properties

3. ODUflex in OTN:

Optical Transport Network's flexible containers:

• Arbitrary bandwidth in ~1.25 Gbps increments
• Resize during operation
• Optimized for modern variable-rate demands

Beyond transmission links, circuit switching requires reserving resources within switching systems. The switch fabric—the internal mechanism that connects input ports to output ports—is a critical reservation point.

Switch fabric architectures:

1. Space Division Switching:

Physically separate paths through the switch for each circuit:

• Crossbar switches: N×N matrix of crosspoints
• Multistage switches: Clos networks (input, middle, output stages)
• Each connection uses specific crosspoint(s) exclusively

Crossbar reservation:

         Output Ports
        1   2   3   4
      ┌───┬───┬───┬───┐
    1 │ X │   │   │   │  ← Port 1→1 connected
      ├───┼───┼───┼───┤
I   2 │   │   │ X │   │  ← Port 2→3 connected  
n     ├───┼───┼───┼───┤
p   3 │   │   │   │ X │  ← Port 3→4 connected
u     ├───┼───┼───┼───┤
t   4 │   │ X │   │   │  ← Port 4→2 connected
      └───┴───┴───┴───┘

Each circuit reserves one crosspoint. A 1000×1000 crossbar needs 1,000,000 crosspoints.

2. Time Division Switching:

Shares physical paths using time multiplexing:

• All input samples written to shared memory
• Output reads from memory at appropriate times
• No physical crosspoints—all switching via memory addressing

TDM switch reservation:

Slot 5 Input → Write to Memory Location 205
Slot 12 Output ← Read from Memory Location 205

Connection table:
| Input Slot | Memory Location | Output Slot | Output Trunk |
|------------|-----------------|-------------|---------------|
| 5          | 205             | 12          | Trunk B       |
| 8          | 208             | 3           | Trunk C       |

Reservation = allocating memory location and connection table entry.

3. Combined Time-Space-Time (TST) Switching:

Large switches combine both approaches:

• First stage: Time switching (slot interchange)
• Middle stage: Space switching (crossbar/Clos)
• Final stage: Time switching (slot interchange)

Reservation must secure resources at all stages for a circuit to be established.

Resource reservations are not eternal—they have lifecycles that must be actively managed. Understanding reservation lifetime management is essential for network reliability and efficiency.

Reservation states:

Soft state vs. hard state:

Reservation systems differ in how they maintain state:

Hard state (traditional circuit switching):

• Once established, reservation persists until explicitly released
• No periodic refresh required
• Failure or miscommunication can cause stuck reservations
• Requires explicit cleanup mechanisms

Soft state (RSVP and similar):

• Reservations must be periodically refreshed
• If refresh stops, reservation times out and releases automatically
• More robust to partial failures
• Higher signaling overhead during normal operation

Lifetime management mechanisms:

1. Explicit release:

• Normal teardown sends release messages
• Each node releases resources upon receiving/sending release
• Fast, deterministic, requires working signaling

2. Timeout release:

• If no activity or signaling for extended period, automatic release
• Catches orphaned circuits from signaling failures
• Typical timeout: 30 seconds to several minutes

3. Audit and reconciliation:

• Periodic comparison of circuit tables between adjacent switches
• Circuits that exist on one side but not the other are cleared
• Catches mismatches from transient signaling issues

Resource reservation doesn't happen in isolation—it's intimately connected to traffic engineering, the practice of optimizing how traffic flows through the network. Reservation decisions affect network utilization, and traffic engineering goals constrain reservation choices.

Traffic engineering objectives:

1. Maximize throughput — Carry as much traffic as possible
2. Minimize blocking — Reject as few connection requests as possible
3. Balance load — Avoid overwhelming some links while others are idle
4. Meet SLAs — Honor service level agreements for priority customers
5. Optimize cost — Route traffic on least-cost paths when possible

Reservation strategies:

1. First-fit allocation:

• Allocate the first available resource that meets requirements
• Simple, fast, but can lead to fragmentation
• May leave 'holes' that can't satisfy larger requests

2. Best-fit allocation:

• Allocate the smallest resource that meets requirements
• Minimizes waste per allocation
• Can lead to exhaustion of small resources for small requests

3. Load-balanced allocation:

• Spread allocations across available resources
• Prevents hot spots on individual links or switches
• May use longer paths than necessary

4. Constraint-based routing:

• Find paths that satisfy multiple constraints (bandwidth, delay, reliability)
• Used in MPLS-TE and similar systems
• Compute before allocating to ensure satisfaction

Bandwidth protection and preemption:

Not all circuits are equal. Networks implement priority classes that affect reservation behavior:

Protection classes:

• Unprotected — Resources may be reclaimed if needed by higher priority
• Soft protected — Maintained unless severe congestion
• Hard protected — Never preempted; highest reliability guarantee

Preemption behavior:

When a high-priority request cannot be satisfied:

1. Identify lower-priority circuits on the desired path
2. Release (preempt) enough capacity to satisfy the request
3. Affected circuits are torn down or rerouted
4. High-priority circuit is established

Preemption ensures critical services (emergency calls, priority customers) can always get capacity, but requires careful policy design to avoid abuse.

While traditional circuit switching used implicit reservation (connection establishment = resource allocation), modern networks implement explicit reservation protocols. These bring circuit-switching-like guarantees to packet-based infrastructure.

RSVP (Resource Reservation Protocol):

RSVP enables end hosts to request specific QoS from the network:

How it works:

1. Sender sends PATH messages describing traffic characteristics
2. PATH messages propagate to receiver, establishing path state
3. Receiver sends RESV message back, requesting capacity
4. Each router along path allocates resources based on RESV
5. Traffic flows with reserved capacity

Key RSVP concepts:

• Flow specification (flowspec) — Describes the traffic (rate, burst size)
• Filter specification (filterspec) — Identifies packets belonging to the flow
• Soft state — Reservations require periodic refresh
• Receiver-driven — Receivers determine reservation level

MPLS-TE (Traffic Engineering):

MPLS-TE creates reserved paths through an MPLS network:

Signaling options:

• RSVP-TE — Extended RSVP for MPLS label binding
• CR-LDP — Constraint-based LDP (largely obsoleted by RSVP-TE)

Capabilities:

• Explicit path specification — Administrator defines exact route
• Bandwidth reservation — Reserve capacity on each link
• Fast reroute — Pre-computed backup paths for failure recovery
• Constraint-based routing — Automatic path selection meeting constraints

Segment Routing:

The latest evolution simplifies reservation:

• Path encoded in packet header as segment list
• No per-flow state in network core
• Bandwidth can be reserved via external controllers
• Combines circuit-like paths with packet-switching flexibility

Pure circuit switching reserves resources conservatively—if a connection might need capacity, that capacity is fully reserved. But most traffic is bursty; circuits are often idle. Oversubscription and statistical multiplexing exploit this to improve efficiency.

The insight:

Not all reserved circuits are active simultaneously. If we reserve 100 Mbps for 10 users each 'needing' 10 Mbps, but they're only active 20% of the time on average:

• Peak demand (everyone active): 100 Mbps
• Average demand: 10 × 10 × 20% = 20 Mbps
• 80 Mbps sits idle under full reservation

Statistical multiplexing shares that 100 Mbps among users, betting that not everyone peaks simultaneously.

Oversubscription ratio:

The oversubscription ratio is total subscribed capacity divided by actual capacity:

Oversubscription Ratio = Total Subscribed Capacity / Physical Capacity

Example:
- Physical capacity: 100 Mbps
- 50 users each with 10 Mbps 'guaranteed'
- Total subscribed: 500 Mbps
- Ratio: 500 / 100 = 5:1

Common oversubscription ratios:

Risks of oversubscription:

When statistical assumptions fail:

• Congestion events — Many users active simultaneously
• Quality degradation — Packets dropped or delayed
• Broken guarantees — Promised capacity unavailable

Managing oversubscription:

1. Traffic shaping — Enforce rate limits even when capacity exists
2. Priority queuing — Favor premium traffic during congestion
3. Admission control — Refuse new flows when oversubscribed
4. Burst allowances — Permit short bursts, constrain sustained rates

We have comprehensively explored how resources are reserved in circuit-switched networks. Let's consolidate the essential insights:

What's next:

Now that we understand resource reservation mechanics, we'll examine the PSTN (Public Switched Telephone Network)—the ultimate example of circuit switching at global scale. You'll see how dedicated paths, connection establishment, and resource reservation combine to create the most reliable communication network ever built.