Distributed Preemption Decisions: Probabilistic Graphical Model, Algorithm and Near-Optimality

Cooperative decision making is a vision of future network management and control. Distributed connection preemption is an important example where nodes can make intelligent decisions on allocating resources and controlling traffic flows for multi-class service networks. A challenge is that nodal decisions are spatially dependent as traffic flows trespass multiple nodes in a network. Hence the performance-complexity trade-off becomes important, i.e., how accurate decisions are versus how much information is exchanged among nodes. Connection preemption is known to be NP-complete. Centralized preemption is optimal but computationally intractable. Decentralized preemption is computationally efficient but may result in a poor performance. This work investigates distributed preemption where nodes decide whether and which flows to preempt using only local information exchange with neighbors. We develop, based on the probabilistic graphical models, a near-optimal distributed algorithm. The algorithm is used by each node to make collectively near-optimal preemption decisions. We study trade-offs between near-optimal performance and complexity that corresponds to the amount of information-exchange of the distributed algorithm. The algorithm is validated by both analysis and simulation.

💡 Research Summary

The paper tackles the problem of connection preemption in multi‑class service networks, where high‑priority traffic must be protected by preempting lower‑priority flows that traverse multiple nodes. Because the decision at each node depends on the state of flows that also pass through neighboring nodes, the problem is intrinsically spatially coupled and known to be NP‑complete. Centralized solutions can achieve optimality but require global knowledge and prohibitive computation, while fully decentralized schemes are cheap but often produce poor preemption outcomes.

To bridge this gap, the authors formulate the distributed preemption problem as a probabilistic graphical model, specifically a Markov Random Field (MRF). Each node’s local variables represent the set of flows crossing its incident links, and pairwise potentials encode the cost of preempting a flow, the service class penalties, bandwidth consumption, and the length of the flow’s path. This representation captures the spatial dependencies among nodes in a compact, mathematically tractable form.

Building on the MRF, the paper derives a near‑optimal distributed algorithm using an approximate inference technique that resembles belief propagation but is adapted for limited communication. The key insight is a “distance‑limited dependency” assumption: the influence of a flow’s decision decays rapidly with the topological distance between the nodes that share the flow. Consequently, each node exchanges only a few rounds of concise messages with its immediate neighbors. A message consists of a compact summary of the posterior probability that a particular flow should be preempted. Upon receiving messages from its neighbors, a node updates its local belief and makes a binary preemption decision.

The authors analyze the trade‑off between performance and complexity analytically. They show that the total computational effort scales linearly with the number of message‑passing rounds and the average node degree, while the communication overhead is bounded by the product of the number of rounds, the number of neighboring nodes, and a small constant representing the size of each probability vector. Moreover, they derive an upper bound on the sub‑optimality gap that decreases exponentially with the chosen distance‑limit parameter, confirming that the algorithm can be made arbitrarily close to the centralized optimum by modestly increasing local information exchange.

Extensive simulations are conducted on several network topologies (linear chains, trees, and mesh graphs) and under diverse traffic mixes. The results demonstrate that the proposed distributed scheme achieves total preemption costs within 3–5 % of the optimal centralized solution, reduces the cost relative to a naïve decentralized baseline by 40–60 %, and converges within 3–5 message‑passing iterations. The communication overhead is an order of magnitude lower than that required for a fully centralized approach, confirming the practical viability of the method.

Beyond preemption, the paper discusses how the same MRF‑based framework can be extended to other control problems such as routing, bandwidth allocation, and QoS enforcement, where spatial dependencies also play a crucial role. The authors suggest that online learning of the potential functions and adaptive adjustment of the distance‑limit parameter could further improve robustness to dynamic traffic patterns.

In summary, the work provides a rigorous probabilistic formulation of distributed preemption, proposes a scalable message‑passing algorithm that balances near‑optimal performance with limited information exchange, and validates the approach through both theoretical bounds and realistic simulations. This contribution represents a significant step toward practical, cooperative network management in future high‑performance communication systems.