A slice is a logical network carved out of one physical one, end to end, with its own guarantees. 5G lets a single RAN, transport and core serve a throughput service, a latency-critical service and a massive-sensor service at once — each isolated so a surge on one cannot starve another.
A network slice is a complete logical network — radio, transport and core — carved out of shared physical infrastructure and dedicated to a service. The point is that it runs end to end: slicing only the core, or only the radio, does not deliver a guarantee, because the weakest unsliced link sets the experience.
The canonical three are eMBB for throughput (video, broadband), URLLC for ultra-reliable low latency (control, automation) and mMTC for massive machine-type density (sensors). A device selects a slice with an S-NSSAI — a slice/service type plus an optional differentiator — and the network routes and resources it accordingly.
The reason to slice rather than just prioritise traffic is isolation: each slice is held to its own service-level agreement, and load or failure in one must not degrade another. A stadium filling an eMBB slice cannot be allowed to add a millisecond to a URLLC control loop sharing the same cell.
That isolation has to hold in every domain. In the RAN it means reserving or scheduling radio resources per slice; in the transport it means separation and QoS across the fronthaul and backhaul; in the core it means separate network functions or instances, often with a dedicated user plane function per slice for the latency-sensitive ones.
A low-latency slice usually needs its user plane function pushed to the edge, close to the radio, so traffic breaks out locally instead of crossing the backhaul. A throughput slice may be happy with a central user plane. So slicing and user-plane placement are the same conversation: the slice's SLA decides where its anchor sits.
This is also where slicing meets private networks and neutral host: a campus can run its own URLLC slice for automation alongside an operator's eMBB slice on shared radio, each isolated and each metered against its own agreement.
End-to-end isolation across three domains and often multiple vendors is an orchestration problem, not a feature. Admission control has to refuse a slice it cannot honour; monitoring has to prove the SLA per slice; and the RAN scheduler has to enforce guarantees in real time without stranding capacity the other slices could use.
Over-provision and slicing wastes the spectrum it was meant to use efficiently; under-provision and the guarantees fail under load. Getting that balance right, and proving it, is the real engineering of a sliced network.
Partitioning one physical network — RAN, transport and core — into multiple isolated logical networks, each dedicated to a service with its own throughput, latency, reliability and security, held to its own SLA. A device selects a slice with an S-NSSAI.
The three canonical slice types. eMBB (enhanced mobile broadband) is for throughput; URLLC (ultra-reliable low-latency communication) is for latency-critical control; mMTC (massive machine-type communication) is for huge numbers of low-power sensors.
The identifier a device and network use to select a slice: a Slice/Service Type (SST) plus an optional Slice Differentiator (SD). It tells the network which slice to route and resource a session on.
Because slicing provides isolation, not just priority. Each slice has its own SLA and resources so that load or failure in one cannot degrade another — a guarantee priority alone does not give, especially across the RAN, transport and core together.
A low-latency slice typically needs its user plane function at the edge for local breakout, while a throughput slice can use a central one. The slice's SLA decides where its anchor sits, so slicing and UPF placement are the same decision.
Enforcing end-to-end isolation across the RAN, transport and core, often across multiple vendors. It needs admission control, per-slice SLA monitoring and a real-time scheduler that honours guarantees without stranding capacity — an orchestration problem more than a single feature.
Where a slice's anchor sits.
KNOWLEDGEBands a slice runs on.
KNOWLEDGESharing sliced radio.
KNOWLEDGEEnforcing slice SLAs.
PRODUCTA campus slice on site.
KNOWLEDGEWhen SA unlocks slicing.
End-to-end isolation across RAN, transport and core, with admission control and per-slice monitoring that proves the guarantee, is the real work. Tell us the services and the SLAs.