Concept
Network Slicing (Network Slicing) refers to creating multiple logical network instances on the same 5G infrastructure through virtualization and orchestration. Each slice can be configured with different SLAs, such as ultra-low latency, high bandwidth, or massive connectivity. Terminals access the corresponding slice via selected DNN, S-NSSAI, and other parameters, achieving isolation from the public internet or other service traffic.
And IoT modules
Modules and SIMs need to support slice identifiers assigned by the operator for industry customers; the application side must align with the core network policy; otherwise, it may fall into the default slice. Scenarios such as industrial private networks, V2X, and remote healthcare are often linked to slice agreements, and testing and acceptance should cover latency and packet loss indicators within the slice.
Implementation Challenges
End-to-end slicing requires coordination among wireless, transport, core network, and terminal; misalignment in any part can result in 'slice exists in name, but not in experience.' Projects should clarify responsibility boundaries and fallback strategies (e.g., default public network).
Slice Identifiers and Core Network Policies
On the terminal side, slices are selected via NSSAI, S-NSSAI, etc.; SIM and subscription data must be consistent with the operator's HSS/UDM. QoS flows, GBR, and 5QI map to wireless DRB and core UPF paths, and testing must verify latency percentiles under busy and edge coverage conditions. If the service requires local offloading (MEC), DNS and routing policies must be bound to the slice to avoid traffic looping back to the central cloud.
Testing, Acceptance, and Operations
Acceptance criteria should include attachment success rate, slice selection success rate, end-to-end latency and jitter, packet loss rate, and handover performance. Operations should monitor slice-level KPIs versus the default public network, with automatic alerts for anomalies. Module and SIM batch changes may introduce compatibility issues and should be included in change management. In terms of cost, slices often come with SLA premiums, requiring ROI analysis on business benefits and redundant links.
Engineering Practice Supplement
Combine '5G network slicing' with the full lifecycle of the module: verify reference designs and antenna solutions in the EVT stage; conduct conduction/radiation baseline testing and thermal imaging in the DVT stage; solidify calibration and test scripts in the PVT stage. Before certification, check the target country's regulations and operator requirements to avoid missing frequency bands or SAR items. The production line retains golden sample comparisons and sampling rules; after-sales can trace software versions, calibration parameters, and key component batches based on serial numbers. If the customer requests remote operations and maintenance, security upgrade, rollback, and emergency shutdown strategies must also be designed.
Slice identity and subscription data
A network slice is a logical network drawn on the same radio and core equipment. When the terminal registers or opens a session, it carries a slice identity. The core network accepts, replaces, or rejects that identity from the subscription data, then steers the session’s traffic onto the matching user-plane path. Only when the identity matches can quality of service, isolation, and whether traffic passes a local user plane follow it. If the identity does not match, traffic lands on the default slice, and the latency and bandwidth in the contract do not appear by themselves. The module, the SIM, and the radio network must use one set of identities. Changing the SIM data while the module still requests the old value can look like a successful attach on the wrong service path.
The access conditions are complete only when the operator has enabled the slice for that SIM, the radio network in the coverage area advertises the same identity, and the application uses the specified data-network name. Default internet access on the public network does not need these parameters. Do not configure the module as “force this slice” when it is not subscribed, or registration is rejected and the device looks out of service. Acceptance should check, both in the busy hour and at the cell edge, that the session is still on the agreed path. One unloaded measurement at a near point does not show that the slice is in effect.
Local breakout and a DNS result that disagree
The user-plane function sits in the campus, but the domain name still resolves to the central machine room. Traffic goes out and comes back, and the slice’s latency figure is already broken on the path. The default slice and the industry slice share one access-point name, and the module activates only the first session. A handover into a neighbor cell that does not deploy the slice has no fallback, so the service stops instead of returning to an acceptable public path. The subscription identity on the test SIM differs from the production SIM, and the lab conclusion fails entirely on site. Treating the instantaneous radio rate as the slice’s end-to-end figure leaves out queuing in the core and on the transport. After the slice contract changes, the old identity is still in the module configuration, and it cannot be corrected remotely outside an upgrade window.
Design the domain name and the route together with the slice. The lookup result should land on an address the current user plane can reach. Address family and prefix changes are covered in cellular IPv6. If the resolved server allows only a connection opened by the device, the direction in NAT and private networks still applies. Slice isolation does not by itself let the cloud connect straight to the terminal’s private address. Write the fallback down: when the requested slice is unavailable, either refuse the service or allow a downgrade to the default path. The two cannot be decided verbally on site.
At registration, check that the slice identity the module requests and the SIM subscription are the same set. If they differ, do not read a successful attach as “already on the industry path.” The DNS result has to land on an address the current user plane can reach, or traffic detours to the center and back. When a neighbor cell does not have the slice, refuse the service or downgrade according to the policy written in advance. Do not decide on site. Check both the busy hour and edge coverage to see whether the session is still on the agreed path. An unloaded rate at a near point does not stand in for end-to-end latency. After a SIM change, read the subscription data again. A conclusion from the test SIM cannot be carried straight onto the production SIM.