Core Placement Policies in NFV & MEC

• Core Placement Policies are principles and algorithms that map logical network functions to physical resources, balancing latency, bandwidth, and resource constraints.
• They leverage integer linear programming and selective replication of data-plane VNFs to optimize resource utilization and ensure quality of service.
• Adopting these policies enhances operational scalability and efficiency in NFV and MEC deployments by minimizing delay and reducing unnecessary overprovisioning.

Core placement policies define principles and algorithms for mapping logically distinct network, software, or hardware "core" components onto physical resources in distributed, virtualized, or multi-core computing systems. In classical and emerging network contexts, including NFV-based telco cores and multi-access edge environments, the placement of core functions has direct and quantifiable effects on bandwidth usage, signaling and data-plane latency, application QoS, resource utilization, and operational scalability. The trade-offs and constraints surrounding these placements require rigorous mathematical modeling, state-aware replication, and a comprehensive balancing of network, compute, and application-layer requirements.

1. Architectural Models and Service Chains

The Evolved Packet Core (EPC) of mobile networks has historically been deployed as a centralized architecture. Virtualization via Network Function Virtualization (NFV) transforms each core function into a Virtual Network Function (VNF), now subject to dynamic, distributed placement on commodity NFV-capable hardware. The system distinguishes between control plane chains—modeled as "Control Service Chains" (CSCs) reflecting ordered Non-Access Stratum (NAS) interactions (e.g., attach, bearer setup, mobility events)—and data plane chains ("Data Service Chains", DSCs) that connect UEs to gateways or services post-authentication.

A fundamental property of these chains is that core functions are stateful: each VNF instance manages session- or flow-specific state, driving the need for instance-tracking when mapping logical service chains to physical resource paths. With the rise of Multi-Access Edge Computing (MEC), core placement must also account for proximity to users to minimize backhaul and application delay, favoring the distribution of certain functions.

2. Integer Linear Programming Model and Constraints

Optimal core placement is cast as an integer linear programming (ILP) problem, with an objective to minimize aggregate network-resource (bandwidth) consumption subject to latency and resource constraints. The objective is given as:

$$\min \sum_{c \in \mathcal{C}} \sum_{\ell \in L} \sum_{i=1}^{n_c-1} D^c \beta_i^c y_{i\ell}^c,$$

where $D^c$ is the demand for service chain $c$, $\beta_i^c$ the fraction of demand between VNF indices $i$ and $i+1$, and $y_{i\ell}^c$ a binary variable indicating use of link $\ell$ for this sub-path.

Latency requirements are enforced via:

$$\sum_{\ell \in L} \sum_{i=1}^{n_c-1} \Delta_p(\ell) y_{i\ell}^c + \sum_{i=1}^{n_c} \beta_{i-1}^c D^c \Delta_{\text{proc}}^{(f_i)} \leq L^c$$

for each chain $c$, where $\Delta_p(\ell)$ is per-link propagation latency and $\Delta_{\text{proc}}^{(f_i)}$ the per-function processing latency. Node and link capacity constraints, for example in CPU cores per Gbps for NFV nodes, are explicitly captured:

$$\sum_{c \in \mathcal{C}} D^c \sum_{i,j: f(f_i^c)=f} \beta_{i-1}^c (\text{cores per Gbps for } f) x_{v,i}^c \leq \text{Total cores at node } v$$

Bandwidth constraints similarly bound flows on physical links.

3. Distributed vs. Centralized Deployment and Selective Replication

The core policies identify a critical dichotomy: distributed versus centralized deployment. Analysis demonstrates that distributing vEPC functions—especially data-plane elements (SGW, PGW)—closer to aggregation points and MEC nodes reduces average end-to-end path lengths, yielding significant decreases in total bandwidth consumption and propagation latency.

However, not all functions require distribution. Empirical results validate that replicating only the data-plane VNFs achieves nearly the same bandwidth and delay gains as a blanket replication of all core functions. Control-plane components (MME, HSS, PCRF), generally characterized by lower signaling volume and higher latency tolerance, can remain centralized or minimally replicated without incurring a major performance penalty. This selective replication paradigm materially simplifies state consistency and operational overhead while preserving network-resource efficiency.

The placement policy thus operates with a minimal-replica principle: beyond two distributed replicas, incremental returns in efficiency rapidly diminish.

4. Implications for Bandwidth, Delay, and Scalability

The principal impact of optimal core placement is a measurable reduction in network-resource utilization and application-layer delay. Simulation and ILP results show a marked reduction in bandwidth consumption when shifting from a centralized to a minimally distributed deployment. Integrating MEC, with application functions placed at the edge, further augments these savings by localizing traffic, thereby averting unnecessary core traversal.

Latency constraints embedded in the ILP ensure that both signaling and user-plane delays satisfy the stringent requirements of contemporary applications. By explicitly modeling both propagation and function processing times, and distributing replicas of high-impact VNFs to strategic sites, the policy meets application QoS without excessive overprovisioning.

Scalability is enhanced by supporting selective, demand-driven replication—facilitating transparent scaling of critical VNFs (primarily SGW/PGW), while constraining the proliferation of services which would otherwise complicate consistency and state synchronization in a fully distributed deployment. The ILP model is inherently adaptable, permitting rapid recalculation in response to evolving demand, topologies, or application latency targets.

5. Operationalization and Broader Context

Implementing these placement policies requires an NFV infrastructure that supports flexible instantiation and migration of VNFs, tied into an orchestration framework capable of enforcing ILP-derived placements. Key operational levers include the ability to statefully replicate core functions, dynamic link and node monitoring for capacity enforcement, and integrated support for MEC hosting.

From a broader network management perspective, these core placement policies provide foundational guidance for evolving metro-scale networks, especially in the context of 5G/6G architectures, where proximity, low latency, and cost minimization are paramount.

The approach also generalizes to other distributed function placement problems where stateful, interdependent service chains must be efficiently mapped to constrained, geo-distributed infrastructures.

6. Summary Table: Optimal Core Placement Policy Features

Policy Dimension	Central Finding	Implication
Control/Data Distinction	Only data-plane VNFs require replication	Simplifies deployment, reduces state overhead
Distributed Placement	Major gains achieved with 2+ replicas	Beyond 2, marginal benefits diminish
Latency/Bandwidth Trade	Distributed placement cuts both	Ensures QoS and reduces resource use
Resource Constraints	Explicitly modeled in ILP	Prevents node/link overload
Integration with MEC	Edge hosting further reduces delay/load	Supports emerging applications

7. Concluding Perspectives

Core placement policies, as formalized through constrained optimization and validated by simulation, enable next-generation networks to reconcile the demands of application QoS, network efficiency, and operational scalability. The findings outlined, notably the sufficiency of limited, selective data-plane VNF replication and the minor incremental benefit of full distribution, establish a practical methodology for deploying scalable, resource-efficient NFV-based mobile core architectures while supporting advanced multi-access edge and low-latency services (Gupta et al., 2018).