Network Slicing: Partitioning a Single Physical Network

Network slicing allows an operator to create multiple logically isolated end-to-end networks on a single physical infrastructure. Each slice is tailored for a specific service type with guaranteed performance characteristics — from bandwidth and latency to reliability and device density.

The concept is standardized in TS 23.501 Section 5.15 and the procedures for slice selection are defined in TS 23.502 Section 4.2.2.2.2. Slicing spans all three network domains: RAN, transport, and core, enabling true end-to-end service differentiation.

S-NSSAI: The Slice Identifier

Every network slice is identified by a Single Network Slice Selection Assistance Information (S-NSSAI), which consists of two fields defined in TS 23.003 Section 28.4.2:

• SST (Slice/Service Type): An 8-bit field (0–255) identifying the expected behavior of the slice
• SD (Slice Differentiator): An optional 24-bit field for further differentiation within the same SST

Standardized SST Values

The UE includes the Requested NSSAI (a list of up to 8 S-NSSAIs) in the NAS Registration Request. The AMF, with assistance from the NSSF, determines the Allowed NSSAI — the set of slices the UE is permitted to use on this PLMN.

NSSAI Hierarchy

Slice Type Comparison: Performance Characteristics

Each slice type targets different KPIs. The table below compares the four primary slice types across critical performance metrics.

End-to-End Slice Architecture

A complete network slice spans three domains, each with dedicated or shared resources:

RAN Slicing

The gNB supports multiple slices simultaneously through RAN slicing mechanisms defined in TS 38.300 Section 16.3:

• MAC scheduler partitioning: Dedicated PRB allocation per slice (e.g., 40% for eMBB, 10% reserved for URLLC, 50% shared)
• QoS flow mapping: Each slice's QoS flows are mapped to dedicated Data Radio Bearers (DRBs)
• Slice-aware admission control: The gNB rejects connections if a slice's resource quota is exhausted
• Pre-emption for URLLC: URLLC traffic can pre-empt eMBB allocations when latency budgets are at risk (per TS 38.213 Section 11.2)

Transport Slicing

The transport network between RAN and Core uses:

• MPLS/Segment Routing: Traffic engineered paths with per-slice bandwidth guarantees
• VLAN/VxLAN tagging: Slice identifier mapped to transport encapsulation
• Typical SLA enforcement: Per-slice committed information rate (CIR) and peak information rate (PIR)

Core Slicing

Each slice can have dedicated or shared NF instances:

• Dedicated AMF: For high-security enterprise slices requiring complete control-plane isolation
• Shared AMF + dedicated SMF/UPF: Most common deployment; AMF handles multi-slice UEs while each slice gets its own session management and user plane
• Dedicated UPF at edge: For URLLC slices requiring local breakout

Worked Example: Enterprise Slice Dimensioning

A manufacturing company requires a private URLLC slice for 200 AGVs (Automated Guided Vehicles) in a factory:

Requirements:

• Latency: <5 ms E2E
• Reliability: 99.999%
• Throughput per AGV: 10 Mbps UL, 5 Mbps DL
• S-NSSAI: SST=2, SD=0x000A01 (operator-assigned)

RAN allocation: `

Total UL demand: 200 AGVs × 10 Mbps = 2 Gbps

Required PRBs (n78, 30 kHz SCS, QPSK for reliability):

Each PRB delivers ~168 kbps UL (QPSK, 2 layers)

PRBs needed: 2,000,000 / 168 = ~11,905 PRBs/s

Per slot (0.5 ms): 11,905 / 2,000 = ~6 PRBs per slot

This represents ~2.2% of 273 available RBs per slot

` Core allocation: `

Dedicated edge UPF: 2 vCPUs, 4 GB RAM (handles 200 sessions)

Shared SMF: Slice-specific DNN "agv.factory.local"

Dedicated N6 path: Direct fiber to factory MEC server

` Transport: Dedicated MPLS LSP with CIR = 2.5 Gbps, latency budget <2 ms for transport segment.

NSSF: The Slice Selection Brain

The Network Slice Selection Function (NSSF) plays a critical role during registration, as defined in TS 29.531:

1. Receives the Requested NSSAI from the AMF
2. Determines the Allowed NSSAI based on subscription data, operator policy, and current slice availability
3. If the serving AMF cannot serve the requested slices, NSSF recommends a target AMF or AMF Set that supports them
4. Returns the mapping between Allowed NSSAI and AMF assignments

The NSSF maintains a mapping table of S-NSSAI to NF instance sets, enabling load-aware slice assignment.

Real Operator Deployments

SK Telecom — Commercial 5G Slicing (South Korea)

SK Telecom launched commercial network slicing in Q3 2023, among the first operators globally:

• Three production slices: Premium eMBB (SST=1, SD=premium), Standard eMBB (SST=1, SD=standard), Enterprise URLLC (SST=2)
• Premium slice SLA: Guaranteed 1 Gbps DL and <10 ms latency, priced at a 30% premium over standard plans
• Implementation: Dedicated SMF/UPF instances per slice, shared AMF across slices, Samsung and Ericsson RAN with scheduler partitioning
• Results: Premium slice users reported 3x median throughput improvement during congestion periods at stadiums and concert venues
• n78 deployment with 100 MHz bandwidth partitioned: 60% eMBB standard, 30% eMBB premium, 10% URLLC reserved

KDDI — Enterprise Slicing (Japan)

KDDI deployed enterprise network slicing in partnership with Samsung in 2024:

• Target: Smart factory deployments for Toyota and Denso manufacturing facilities
• Slice configuration: SST=2 (URLLC) with customer-specific SD values per factory
• Architecture: Dedicated UPF co-located at factory premises (on-prem MEC), shared SMF in regional DC
• Performance achieved: 2.1 ms average E2E latency (target was <5 ms), 99.9997% reliability over 6-month observation
• Scale: 500+ connected industrial robots per factory, each requiring 15 Mbps UL for vision data

Worked Example: Slice Revenue Modeling

An operator with 10 million subscribers models slicing revenue:

Annual revenue uplift: $249.6M — representing a ~7% increase on assumed base ARPU of $30/mo.

Slice Isolation Mechanisms

True slice isolation requires enforcement at multiple layers, as outlined in TS 33.501 Section 16:

• Resource isolation: Dedicated compute, memory, and storage for critical NFs (hard isolation)
• Traffic isolation: Separate GTP-U tunnels per slice on N3/N9 interfaces
• Security isolation: Per-slice authentication possible via NSSAA (Network Slice-Specific Authentication and Authorization) defined in TS 33.501 Section 16.2
• Failure isolation: Fault in one slice's UPF does not impact other slices' UPFs
• Management isolation: Separate FCAPS (Fault, Configuration, Accounting, Performance, Security) per slice via dedicated NSSMF (Network Slice Subnet Management Function)

Key Takeaway: Network slicing transforms 5G from a single-purpose pipe into a multi-service platform. The S-NSSAI framework (SST + SD) provides a clean addressing mechanism, and the interplay between NSSF, AMF, and dedicated NF instances enables true end-to-end isolation. Real deployments from SK Telecom and KDDI demonstrate that slicing is no longer theoretical — it is generating revenue and enabling use cases that were impossible with previous network generations. Engineers must understand slice dimensioning, isolation requirements, and NSSAI procedures to design production-grade sliced networks.