How a 4G Network Adds a New MME Without Disrupting Existing Ones (Part 1: Weight Factors)

Imagine you run a live 4G network and need to add a new MME to a pool. Millions of devices are already attached to the existing MMEs. How do you get the new node to take its fair share of traffic without moving or disturbing anyone?
3GPP’s answer for new arrivals is simple: the Weight Factor.

The idea: probability proportional to weight

Each MME in a pool is assigned a Weight Factor, typically based on its capacity relative to the other MMEs. When a UE enters the MME Pool Area, the eNodeB selects an MME with a probability proportional to that MME’s Weight Factor. No live load measurement is needed, just a configured number.

A worked example (illustrative numbers)

Suppose a pool has three MMEs with Weight Factors of 10, 10 and 20. The total is 40, so new UEs are directed to them with probabilities of 25%, 25% and 50%. The largest MME takes about half of the new arrivals, which matches its larger capacity.

Now the operator installs a fourth MME. It may be given a much higher Weight Factor for an initial period, so that its load builds up faster. Once it is carrying its intended share, the weight can be lowered to reflect its real capacity.

The subtle part: only new selections are affected

Changing a Weight Factor changes how the eNodeB selects an MME for UEs that subsequently enter the pool. UEs already registered are not moved, so the load shifts gradually, as new UEs arrive and existing ones leave. (This is my interpretation of how the mechanism behaves, not wording from the spec.)

This also fits with the spec’s intention that the Weight Factor is not changed frequently. In a mature network, monthly changes could be anticipated, for example when RAN or CN nodes are added. The value reflects configured capacity, not short-term traffic, so there is no need to chase it.

How the eNodeB learns the weights

The MME sends its Weight Factor to the eNodeB through S1-AP messages (TS 36.413). If a HeNB GW is deployed, the MME sends it to the HeNB GW instead. The value is delivered over the node-level S1 connection between the eNodeB and the MME, and an operator can change it later without re-establishing that connection.

Extension: Dedicated Core Networks (DCNs)

With DCNs, the same idea is applied per DCN. The eNodeB balances load only among MMEs of the same DCN within the same pool area (same PLMN and MMEGI). An MME serving several DCNs may have a separate Weight Factor for each one, set according to its capacity for that specific DCN. This matters because an MME may have plenty of capacity for one DCN and little for another, so a single weight would distribute one of them badly (my reasoning, not spec text).

A note on low access priority UEs

In some networks, the eNodeB can be configured to select specific MMEs for UEs configured for low access priority, with a different load balance from other UEs. It recognises such UEs from the RRC establishment or RRC resume signalling. As a last resort, when no other information or configuration is available for choosing an MME from a specific DCN, the eNodeB may select an MME from a specific CN for these UEs.

The takeaway

Weight Factors are a lightweight, configuration-driven way to share load. Assign each MME a number, let the eNodeB choose in proportion, and let the load settle as devices come and go. Weight Factors on their own never move a device that is already attached, which is both their strength (no disruption) and their limit (slow to correct an imbalance).

Next in this series: what happens when you need to move already-attached UEs, for example when an MME is being taken out of service or is overloaded? That is load rebalancing, in Part 2.
Have you seen weights tuned in a real network rollout? Please share your views

#4G #LTE #3GPP #Telecom #EPC #MME #LoadBalancing #MobileNetworks

Sources: 3GPP TS 23.401 and TS 36.413

Leave a Reply

Your email address will not be published. Required fields are marked *