Definition
QoS (Quality of Service) uses scheduling, congestion control, prioritization, packet loss strategies, etc., to provide differentiated guarantees of bandwidth, latency, and packet loss for different services. LTE identifies service types through QCI (QoS Class Identifier), 5G through 5QI (5G QoS Identifier), combined with GBR (Guaranteed Bit Rate) or Non-GBR bearer, achieving end-to-end QoS from the air interface to the core network. QoS works collaboratively in traffic scheduling, differentiated services, bandwidth management, and other processes.
Key Parameters and QCI/5QI
The core parameters of QoS include: resource type (GBR/Non-GBR), priority (1–9, with 1 being the highest), packet delay budget (PDB), packet error rate (PER), maximum data burst volume (MDBV), etc. LTE QCI 1–4 are GBR (voice, video calls), QCI 5–9 are Non-GBR (web, downloads, IoT). 5G 5QI extends more service types, such as V2X, industrial control, mMTC, etc., supporting finer-grained latency and reliability requirements.
End-to-end QoS implementation
When a terminal initiates a service, the core network establishes the corresponding QoS EPS Bearer/PDU Session based on APN, DNN, slice, and other subscription information. The air interface scheduler allocates resources by QCI/5QI, while the core network marks headers with DSCP for differentiated forwarding. In IoT scenarios, NB-IoT, Cat.1, etc., can configure low-priority QCI to reduce the impact on high-value services, while critical commands can request high-priority bearers.
SimtoLink's IoT SIM cards and cloud platform can configure different QoS levels based on APN when establishing data connections. For low-rate services such as meters and sensors, Non-GBR bearers are used; for latency-sensitive scenarios such as vehicles and industrial control, higher-priority QCI can be requested. On the SimtoLink platform, in scenarios like command issuance and OTA upgrades, QoS policies are combined to ensure the timeliness and reliability of critical operations.
Guaranteed Bit Rate and Statistical Multiplexing Are Not the Same Bearer
A QoS mechanism writes the expectation for a bearer as parameters a scheduler can carry out. A guaranteed-bit-rate bearer reserves resources at admission. If the network judges that resources are not enough, it refuses the setup instead of accepting it and then doing its best. A non-guaranteed bearer joins statistical multiplexing. It can run fast while the network is idle, and it is the first to be squeezed under congestion. A QCI or 5QI binds resource type, priority, packet delay budget, and packet error rate together. The packet delay budget is the upper target that scheduling and queuing hope to keep. The packet error rate is the residual-error target the radio link is meant to reach. Neither is a measurement promise for the public path after the cellular network. The maximum data burst reminds the scheduler how large a burst may enter at once, so a bearer that is nominally low-latency is not blocked by one large packet.
End to end, every hop has to recognize this classification. The air interface schedules by bearer, the core forwards by DSCP or service class, and the enterprise network queues again. Clear the mark at any hop and later devices can only forward the packet as ordinary traffic. QoS does not add fiber bandwidth. It only changes how the loss is shared when there is not enough.
The Delay Budget Is a Scheduling Target
The condition is that the service can be split onto different bearers or different marks. If meter reports and emergency control share the same default bearer, the scheduler has no second object to prefer. Apply for a guaranteed bit rate only for voice or a control loop that needs a stable delay. Occasional reports that can be retransmitted can use a non-guaranteed bit rate. In 5G, also check whether the QoS flow of the PDU session was actually established. Giving a packet a priority name at the application layer does not create a new QoS flow.
When the packet delay budget is shorter than the radio scheduling period, or shorter than the time to wake for paging, the target cannot be met. An idle device has to be paged and then complete random access. That time is not inside the bearer delay budget. A low-latency service should stay connected, or it should accept the wake time. Across an enterprise private network, a cellular 5QI does not automatically become an intranet queue. Map priority QoS to an agreed DSCP at the exit, and keep the matching queue on the enterprise equipment.
Marked Only on the Air Interface, Rewritten at the Exit
A common mistake is to set QoS on the module and capture at a point where NAT or a tunnel device has already cleared the mark, then conclude that “QoS did not take effect.” Look in three places at once: bearer-setup signaling, air-interface scheduling, and the IP header at the exit. Another mistake is to read the packet delay budget as a contract for delay over the whole path, and to complain about radio parameters when a public-network detour or slow server processing is the cause. Assigning the highest guaranteed bit rate to every IoT SIM fills cell resources at admission, and later voice and alarms fail to be established.
A non-guaranteed bearer is fast in an idle speed test, so acceptance passes, and delay rises at the traffic peak. That is normal statistical multiplexing, not a parameter rollback. Compare loss and queuing delay of different bearers under congestion. If two bearers measure exactly the same, check first whether they actually landed on the same QCI.
While the network is idle, a non-guaranteed bearer can run quickly, and one speed test can mislead acceptance. Create congestion on purpose in the same cell, then compare queuing delay and loss for the alarm flow and the bulk flow. If the two curves lie on top of each other, the flows did not land in different quality classes. If the delay budget is shorter than the time to wake the device from idle, the budget is already used up before paging finishes. Keep that service connected. Do not only change core-network parameters. When the exit mark is rewritten, the air interface is already distinguishing the flows and the enterprise gateway still treats them alike. Compare the class identifier in the bearer-setup signaling and the mark in the exit packet header at the same time.