Core Network Slicing
How a Single 5G Core Becomes Many Virtual Networks
Network slicing is the architectural pillar that lets a 5G operator sell one physical network as many tailored products. In the 4G evolved packet core, every subscriber traversed essentially the same monolithic gateway chain, so a latency-sensitive control message and a bulk video download competed for the same scheduling and buffering resources. The 5G core (5GC) breaks this apart. Because every network function, the AMF, SMF, UPF, PCF, and the rest, is implemented as a stateless microservice on a virtualized or containerized platform, the operator can instantiate a fresh set of those functions for each slice and stitch them into an independent Network Slice Instance (NSI). A device that registers to the network presents a Requested NSSAI, the Network Slice Selection Function maps it to the right NSI, and from that moment its traffic is steered through user-plane functions reserved for that slice.
What makes slicing genuinely end-to-end is that the same partitioning is mirrored in the transport network and the radio access network. A URLLC slice is useless if the core guarantees 1 ms processing but the fronthaul adds 10 ms of jitter, so segment-routing policies reserve bounded-latency paths between the central unit, distributed unit, and core sites, while RAN slicing pins dedicated physical resource blocks (PRBs) or a shorter numerology to that slice. At millimeter-wave frequencies the radio layer carries enormous instantaneous capacity, which is precisely what allows several demanding slices to share one carrier; the wide bandwidth available in the FR2 bands (24.25 to 71 GHz) gives the scheduler the headroom to honor concurrent eMBB and URLLC commitments on the same site.
Each slice is bound to a Service Level Agreement that translates abstract service intent into measurable key performance indicators: a peak and guaranteed bit rate, a one-way latency bound, a packet error rate, a reliability percentile, and a maximum supported device density. The NSSF, the slice manager, and the per-slice admission-control logic together enforce these numbers, rejecting new sessions on a congested slice rather than letting overload bleed across slice boundaries.
Slice Identity and Selection Signaling
A slice is named by its S-NSSAI, which combines an 8-bit Slice/Service Type with an optional 24-bit Slice Differentiator. The Slice/Service Type carries standardized meaning (1 for eMBB, 2 for URLLC, 3 for mMTC, 4 for V2X), while the Slice Differentiator distinguishes multiple tenants or use cases that share the same service type, for example two separate enterprise URLLC slices. A UE may carry a Configured NSSAI of up to eight S-NSSAIs, request a subset at registration, and receive an Allowed NSSAI scoped to the current registration area.
Per-Slice Resource and SLA Mathematics
S-NSSAI = SST (8 bits) [ + SD (24 bits) ] → up to 32 bits total
Guaranteed slice throughput (RAN allocation):
Rslice ≈ NPRB × 12 × (Nsym / Tslot) × Qm × Rcode × η
End-to-end latency budget:
Te2e = TRAN + Ttransport + TUPF + TDN ≤ SLAlatency
Where NPRB = physical resource blocks reserved for the slice, 12 = subcarriers per PRB, Nsym/Tslot = symbols per slot duration, Qm = modulation order bits, Rcode = code rate, η = scheduler efficiency. Example URLLC target: TRAN ≈ 0.5 ms, Ttransport ≈ 0.2 ms, TUPF ≈ 0.2 ms → Te2e ≈ 0.9 ms ≤ 1 ms SLA.
Standardized Slice Profiles
| Slice Type | SST Value | Target Latency | Reliability | Peak Data Rate | Representative Use |
|---|---|---|---|---|---|
| eMBB | 1 | 4 ms (user plane) | 99.9% | 20 Gbps DL / 10 Gbps UL | 4K/8K video, fixed wireless |
| URLLC | 2 | 1 ms (one-way) | 99.999% | Up to ~1 Gbps | Factory automation, remote control |
| mMTC | 3 | Seconds (delay-tolerant) | 99% | Low (kbps per device) | 1 million devices/km² IoT |
| V2X | 4 | 3 to 10 ms | 99.99% | Tens of Mbps | Vehicle platooning, sensor sharing |
| Custom enterprise | 1 to 3 + SD | SLA-defined | SLA-defined | SLA-defined | Private 5G, network-as-a-service |
Frequently Asked Questions
What is the difference between an S-NSSAI and an NSSAI?
An S-NSSAI names one slice: an 8-bit Slice/Service Type (SST) plus an optional 24-bit Slice Differentiator (SD), up to 32 bits total. Standard SST values are 1 (eMBB), 2 (URLLC), 3 (mMTC), 4 (V2X). An NSSAI is the set of S-NSSAIs a device may use; a UE can request up to eight, and the network returns an Allowed NSSAI. The NSSF maps requested S-NSSAIs to Network Slice Instances and selects the serving AMF.
How does the 5G core select which network slice a device connects to?
At registration the UE sends a Requested NSSAI in RRC and NAS signaling. If the initial AMF cannot serve every requested slice, it queries the NSSF, which checks subscribed S-NSSAIs, operator policy, and per-tracking-area availability, then returns the Allowed NSSAI, a Configured NSSAI, and the target AMF set. PDU sessions are then set up toward the slice's dedicated SMF and UPF, so one device can run an eMBB slice and a URLLC slice at once.
What level of isolation do 5G network slices actually provide?
Isolation works across three planes. The core can deploy dedicated SMF and UPF containers per slice so a fault in one will not starve another. The transport network reserves bandwidth and bounds latency via segment routing or IETF slicing. The RAN allocates dedicated PRBs, schedulers, or numerologies. Hard isolation dedicates compute, transport, and spectrum; soft isolation shares functions under per-slice quotas and admission control. The achievable degree depends on what the operator physically dedicates.