LyndHub
Jadwal Sholat

Memuat jadwal sholat…

← All journalsSearch index

Ilmu Komputer & AI editorial

Open AccessOA2026

Centrality-Based Deployment of Queue Policies in Acyclic Multipath Routing Networks

A Katz centrality metric identifies the single best router for Active Queue Management deployment, stabilizing TCP fluid models and preserving low-latency operation under large round-trip times.
Mahima Gupta; Acquin Biju; Rijul Jain; Dipesh Sharma; Sreelakshmi Manjunath· 2026· DOI 10.48550/arXiv.2608.24131

The core problem

Excessive queueing delays remain a primary obstacle to latency-sensitive applications, and while Active Queue Management (AQM) is widely recognized as an effective remedy, its real-world deployment is sparse. The authors address this gap for a specific but practically important class of networks: those in which routers and switches exhibit a topological hierarchy, form acyclic paths, and employ multipath routing. The central premise is that AQM deployment alters packet-forwarding dynamics in networks carrying TCP flows, thereby creating a direct link between network stability and end-to-end performance. The paper asks a concrete design question: given a network with many candidate routers, where should a threshold-based AQM policy be placed to obtain the greatest stabilizing effect? The authors answer this by combining fluid models of TCP and queue dynamics with a graph-theoretic centrality measure, then validating the resulting deployment strategy through packet-level simulations. The work sits at the intersection of network architecture, congestion control, and graph theory, and it offers a lightweight, computable rule for operators who cannot afford to instrument every router.

Innovation

The stability analysis yields two principal findings. First, in the absence of AQM, the network is vulnerable to instability as the average RTT of TCP flows increases; the critical RTT at which the equilibrium loses stability is relatively small. Second, when the threshold-based AQM is deployed at an appropriate router, the network tolerates a much larger RTT without losing stability. The improvement is not marginal: the stability boundary shifts outward substantially, meaning that flows with longer propagation delays can coexist with high link utilization without entering oscillatory or unstable regimes.

The centrality analysis identifies which router qualifies as "appropriate." The authors compute Katz centrality for every node in the acyclic multipath topology and show that the router with the maximum Katz centrality is the one whose AQM deployment produces the greatest stabilizing effect. This is consistent with the intuition that high-centrality nodes carry a disproportionate share of path traffic and therefore exert the strongest influence on the closed-loop dynamics. The result is a simple, topology-aware deployment rule: rank routers by Katz centrality and instrument the t

Excessive queueing delays remain a primary obstacle to latency-sensitive applications, and while Active Queue Management (AQM) is widely recognized as an effective remedy, its real-world deployment is sparse. The authors address this gap for a specific but practically important class of networks: those in which routers and switches exhibit a topological hierarchy, form acyclic paths, and employ multipath routing. The central premise is that AQM deployment alters packet-forwarding dynamics in networks carrying TCP flows, thereby creating a direct link between network stability and end-to-end performance. The paper asks a concrete design question: given a network with many candidate routers, where should a threshold-based AQM policy be placed to obtain the greatest stabilizing effect? The authors answer this by combining fluid models of TCP and queue dynamics with a graph-theoretic centrality measure, then validating the resulting deployment strategy through packet-level simulations. The work sits at the intersection of network architecture, congestion control, and graph theory, and it offers a lightweight, computable rule for operators who cannot afford to instrument every router.
The methodology proceeds in three stages. First, the authors construct a closed-loop fluid model for a network of TCP flows traversing acyclic multipath routes. Each router maintains a queue whose dynamics are governed by a simple threshold-based policy: when the queue length exceeds a threshold, packets are marked or dropped to signal congestion. The TCP window dynamics follow a standard fluid approximation, and the queue length evolves according to the difference between arrival and service rates. For a single congested link, the coupled dynamics can be written schematically as

Why it matters

The paper's contribution is best understood as a bridge between two normally separate literatures: control-theoretic analysis of TCP/AQM dynamics and graph-theoretic analysis of network topology. By expressing the deployment question as a centrality ranking problem, the authors convert a combinatorial search over router subsets into a single scalar computation. This is attractive for operators of hierarchical, acyclic, multipath networks, where the number of candidate routers may be large but the topology is structured enough for centrality measures to be meaningful.

Several limitations and open questions follow from the modeling choices. The fluid model assumes a homogeneous population of TCP flows and a single threshold-based policy; heterogeneous RTTs, non-TCP traffic, and more sophisticated AQM algorithms such as RED or CoDel are not analyzed. The acyclic and multipath assumptions are also restrictive, though they match many data-center and enterprise topologies. The Katz centrality metric depends on a damping factor, and the paper does not fully explore sensitivity to that parameter; in principle, different damping values could reorder the top candidates. Finally, the stability analysis is local, so global behavior under large perturbations remains an open area.

Despite these caveats, the practical message is clear and actionable. For network architects, the result suggests that partial AQM deployment need not be arbitrary: a single well-chosen router, identified by Katz centrality, can deliver most of the stabilizing benefit. This lowers the operational barrier to AQM adoption, which the introduction identifies as a key reason for sparse deployment. The work also suggests a broader research direction: centrality-informed placement of other network functions, such as traffic shaping, load balancing, or security monitoring, where the goal is to maximize a global performance objective with minimal instrumentation. The combination of fluid modeling, stability theory, and graph centrality provides a template that can be adapted to those settings.

Who should read this

CS practitioners and researchers

Opening member content…