Network Slicing
Last updated 13 Oct 2026
At a glance
A network slice is a logical mobile network with its own SLA, realised across RAN, transport and core at once. No single domain controller can deliver it, which makes slicing the use case that justifies cross-domain orchestration. 3GPP management functions and GSMA templates define the contract; the hard part is composing one SLA across domains that change independently.
The picture
One slice, three domains, one end-to-end SLA. Example latency budget of 10 ms.
1 · Slice request
Industrial low-latency slice
- template
- NEST
- S-NSSAI
- SST + SD
- latency
- 10 ms e2e
- isolation
- shared NFs
CSMF turns the customer request into a slice requirement
2 · Subnet slices (NSSMF per domain)
- RAN subnet slice≤ 4 ms
Radio resources, scheduler weighting
- Transport subnet slice≤ 3 ms
SR or RSVP-TE class with matching SLA
- Core subnet slice≤ 3 ms
Slice-aware AMF, SMF, UPF, UDM; NSSF and PCF policy
Each domain evolves on its own cadence; the slice contract must survive it
3 · Instance and assurance
Slice instance
Recorded in inventory as a service with its subnet parts
Slice KPIs armed
On violation: scale UPF, reroute transport, or refuse new attaches
↺ feeds back to the orchestrator
How it works
What a slice is
- A logical mobile network identified by an S-NSSAI (Slice/Service Type plus Slice Differentiator), with its own SLA and policy.
- Realised as a RAN slice (radio resources, scheduler weighting), a transport slice (IP/MPLS or SR with an SLA-aligned class) and a core slice (slice-aware AMF, SMF, UPF, UDM).
- Used for enterprise verticals, IoT, fixed wireless access and low-latency industrial applications.
Standards and templates
3GPP TS 28.530 to 28.541 define slice management and its functions: the CSMF (customer-facing), the NSMF (network slice) and the NSSMF (one per subnet: RAN, transport, core). GSMA GST and NEST standardise slice templates, so an enterprise can ask for a slice in portable terms, such as industrial control with defined latency, jitter and availability. The orchestrator receives a NEST-aligned request, decomposes it into NSMF and NSSMF operations, and tracks the slice instance as a service.
Why slicing breaks simple orchestration
- SLA composition. A 10 ms end-to-end latency slice needs the budget shared between RAN, transport and core. Only the cross-domain layer can guarantee the total.
- Lifecycle independence. RAN, transport and core scale on different cadences, often under different teams or vendors. Each subnet slice must be able to change without breaking the slice contract.
- Isolation or sharing. Strict isolation needs dedicated function instances; shared slicing reuses functions with policy isolation. The catalog encodes the choice and the orchestrator enforces it.
- Assurance feedback. SLA violations must feed back to the orchestrator, which may scale functions, reroute transport or refuse new attaches. Slicing is rarely deployed without integrated assurance.
Worked example: an enterprise slice
Order
COM → SOMAn enterprise orders an industrial low-latency slice. COM and SOM turn it into a NEST-aligned request with latency, jitter and availability targets and an S-NSSAI.
Decompose
Cross-domain orchestratorThe cross-domain orchestrator splits it into a RAN slice configuration, a transport slice request to the IP/MPLS domain, and a core slice across AMF, SMF, UPF and UDM with NSSF and PCF policy.
Scale if needed
NFVO → VNFM → VIMDedicated UPF capacity is requested. MANO scales out a UPF instance; it registers with the NRF and becomes discoverable to slice-aware SMFs.
Configure the core
5GC functionsSlice selection policy, AMF and SMF slice support, slice-bound subscriber data in UDM and slice-specific policy in PCF are applied.
Register and arm assurance
Inventory + assuranceThe slice instance is published to inventory and its KPIs armed in assurance, with closed-loop rules such as scaling the UPF under load or alarming when the latency budget is exceeded.
A full enterprise slice, from order to activation, is worked through in the 5G network slice domain example (11.6).
Who owns what
| Entity | System of Record | System of Engagement | System of Reference | Notes |
|---|---|---|---|---|
| Slice template (NEST / GST-aligned) | Service Catalog | — | NSMF | Encodes SLA targets and the isolation choice. |
| Slice instance | Service Inventory | Cross-domain orchestrator / NSMF | — | TMF638. Tracked as a service with RAN, transport and core subnet parts. |
| Slice order | SOM | — | COM | TMF641 from SOM to the cross-domain layer; state reported back to COM. |
| Subnet slice realisation | Domain NSSMF (RAN, transport, core) | — | Resource Inventory | Each domain owns how its part is delivered; the cross-domain layer owns the budget. |
| Slice SLA performance | Assurance | — | Cross-domain orchestrator | Violations drive closed-loop action. |
In practice
Vendor stacks differ in how much of the slice lifecycle they cover. Nokia’s and Ericsson’s orchestration products span multi-domain workflow and integrate closed-loop assurance, and fit best where that vendor supplies the core and RAN. Mavenir couples its cloud-native core tightly with its own orchestration, common in greenfield and Open RAN builds. ONAP-based stacks give operators code control, with uneven production maturity and heavily customised forks.
The trade-off
What it solves
Slicing lets one physical network carry several logical networks with different SLAs, each ordered and assured as a service.
What it does not solve
- Domains that cannot deliver SLA classes; the slice is only as strong as its weakest subnet.
- Missing slice-level assurance; without it, the SLA cannot be measured or enforced.
- Commercial definition: what a slice is worth to a customer is a product question.
What it adds
- A latency and availability budget split across domains that change independently.
- Closed-loop automation that can scale, reroute or refuse service on its own.
- Isolation choices that multiply the function instances to run and pay for.
Architecture decision
Should slices run on dedicated or shared network functions?
Dedicated function instances
Choose when the customer needs strict isolation or guaranteed performance and pays for the extra capacity.
Shared functions with policy isolation
Choose when many slices with moderate SLAs share capacity, and policy can hold their separation.
Shared by default, dedicated by template
Choose when the catalog offers both, so the isolation level is a product choice and the orchestrator enforces it.
Breaks if the isolation choice lives in engineers’ heads instead of the template. Slices then get whatever the last deployment did, and SLAs cannot be defended.
Takeaways
Key Takeaways
- A slice spans RAN, transport and core; no single domain controller owns it.
- 3GPP CSMF, NSMF and NSSMF and GSMA NEST/GST templates define how slices are requested and managed.
- The cross-domain layer owns the end-to-end budget; each domain owns how its subnet slice is delivered.
- Isolation versus sharing is a catalog decision with a capacity cost.
- Without slice-level assurance and closed loops, a slice SLA cannot be measured or enforced.
Check yourself
The core team says it can deliver a 10 ms slice on its own. Why is that not enough?Show answerHide
The 10 ms is end to end. RAN and transport consume part of the budget, and only the cross-domain layer can allocate and guarantee the total.
A slice’s latency target is breached at peak hours. What should happen, and who decides?Show answerHide
Assurance detects the breach and feeds it back to the orchestrator, which applies the closed-loop rules for that slice: scale the UPF, reroute transport, or refuse new attaches. The rules are set when the slice is designed.