Access Official Technical PDF

Share Summary:LinkedInEmail

Industrial operations depend on networks that connect users, applications, and services across complex infrastructure. Those networks may use stateful firewalls, load balancers, proxies, or other network functions whose decisions depend on traffic seen earlier. Stateful network verification asks whether these devices collectively enforce the intended policies—not just for an individual packet, but as network state changes over time. NetSMC, a research model checker, explores how to make that verification more practical without hiding the assumptions involved.

Why state makes network policies harder to check

In a stateless network, a device’s action can generally be assessed from the packet and its configuration. A stateful network function (NF), by contrast, can consult or update internal tables. A firewall may allow an inbound packet only if a related outbound connection was established earlier. A load balancer may record a server assignment so subsequent packets in a flow reach the same server.

This history dependence makes policy analysis challenging. Operators may need to check isolation, conditional reachability, flow affinity, dynamic service chaining, or path pinning. For example, a policy might require suspicious traffic to visit a heavier inspection stage, or require reverse traffic to pass through the same firewall as the forward flow.

The difficulty is fundamental: even checking a simple isolation property in a general stateful network can be undecidable. A practical verification approach therefore needs explicit trade-offs in the network behavior it models, the policies it supports, or the guarantees it can provide.

NetSMC’s one-packet model

NetSMC reduces the modeled network state to one packet being processed at a time, together with the current state of the network functions. This avoids representing multiple packets moving through the network simultaneously and interleaving at different devices—a major source of complexity.

The simplification is motivated by order-preserving scenarios. Packets that can affect one another’s processing may, in some networks, be handled in the same order by the relevant functions. Connection-based functions provide an example: control packets that establish connection state are often processed before later packets that rely on that state. Under the paper’s stated conditions, the one-packet model can preserve per-packet behavior for a broad set of policies.

That condition matters. A one-packet model cannot reveal violations that arise only from packet interleaving. For policies or networks where interleaving affects outcomes, NetSMC is described as a sound-but-incomplete bug-finding tool: it can identify some violations, but a result that finds no violation does not establish correctness in every case.

Representing network functions and policies

To support efficient analysis, NetSMC uses a structured model of an NF rather than representing its behavior as unrestricted program code. Each function has state tables—key-value maps—and rules that test packet fields or table entries, update state, and forward, modify, or drop packets. The model includes operations such as simple counting and choosing a value from a specified domain.

This structure can represent common behaviors such as a firewall recording established flows or a load balancer storing a client-to-server assignment. It is deliberately constrained: modeling a real function requires expressing its relevant behavior within the supported rules and operations. The paper notes that creating such a model is a one-time task that may be automated, while leaving automation of that process outside its scope.

Policies are expressed in a restricted temporal language based on a subset of linear temporal logic. It supports properties about what happens eventually, what must hold throughout a packet’s traversal, and how behavior should respond when a condition becomes true. Restricting the language makes it possible to use more efficient checking techniques than would be practical for unrestricted temporal specifications.

How symbolic model checking helps

Rather than enumerate every concrete combination of packet values and NF state, NetSMC uses symbolic model checking to represent sets of network states with logical formulas. For example, a formula can describe all states in which a firewall’s connection table contains a permitted flow, without listing every possible table configuration individually.

The checker repeatedly calculates which states can lead to a state that violates a policy, then tests whether an initial state can reach that set. NetSMC’s authors tailor these calculations to the network and policy models. In particular, they connect containment checks between symbolic state sets to query-containment techniques from database theory, rather than relying only on a general-purpose constraint solver.

What the results mean for network operators

The paper reports that NetSMC was evaluated with real network functions including pfSense and HAProxy, across different topologies and policies. It reports that the tool scales to networks with hundreds of stateful NFs and is more than 200 times faster than VMN on typical fat-tree topologies. Those results describe the paper’s evaluation; they do not remove the need to assess whether a deployment fits the model’s assumptions.

For engineering teams, the practical takeaway is to treat formal verification as a model-based process:

  • Define the policy precisely, including the relevant traffic conditions and state transitions.
  • Check that the NF models represent the behaviors that matter to the policy.
  • Determine whether packet interleaving could change the outcome.
  • Interpret a verification result in light of the policy fragment and correctness guarantees supported.

NetSMC illustrates how domain-specific models and symbolic techniques can make stateful network policy checks more tractable. Its value depends on matching the modeled network to the system being assessed and understanding where simplifications apply. For industrial automation decision-makers, that is the central lesson: verification can strengthen confidence in network policy, but its conclusions are only as applicable as its assumptions and models.

Access Official Technical PDF

About The Author