Online Resource Management via Lyapunov Optimization

This direction is merged into another direction: drift-plus-penalty optimization. Its last version stays readable and citable here.

Version 1 of 128 papers34 conceptsfrozen 2026-10-04

In complex networked systems—from maritime edge computing to global federated learning—operators must balance immediate performance (like low latency or high accuracy) against strict long-term resource constraints (like battery life or carbon budgets). Traditional optimization often fails here because it either requires perfect knowledge of the future or makes "myopic" decisions that exhaust resources too early.

The research direction across these papers focuses on Lyapunov optimization, a framework that transforms long-term, time-averaged constraints into a series of stable, real-time decisions. By maintaining "virtual queues" that track resource deficits, these systems can achieve near-optimal performance while strictly adhering to global budgets without ever needing to predict the future.

The Virtual Queue as a Resource Memory

The core intuition behind this approach is the virtual queue. In a physical network, a queue tracks data packets waiting to be processed. In this optimization framework, researchers like Jie Xu and Sheng Zhou treat abstract constraints—such as energy consumption or migration costs—as virtual queues.

If a system has a long-term energy budget, the virtual queue grows whenever the system consumes energy and shrinks at a rate corresponding to the target average budget. As Yongming Huang and Yifan Wang demonstrate in cell-free MIMO networks, this queue acts as a "memory" of past resource usage. When the queue is long, it signals that the system has been "overspending," forcing the algorithm to prioritize conservation in the next time step.

Balancing Performance and Stability

To make decisions, these frameworks use the Drift-plus-Penalty (DPP) algorithm. The goal is to minimize a combination of two factors:

  1. Lyapunov Drift: The change in the virtual queue size. Minimizing this ensures the queue remains stable, which mathematically guarantees that the long-term constraint is satisfied.
  2. Penalty: A cost function representing the performance metric we want to optimize (e.g., latency, training loss, or regret).

By adjusting a control parameter (often denoted as $V$), operators can tune the trade-off. A high $V$ prioritizes performance at the cost of larger virtual queues (and thus more fluctuation in resource use), while a low $V$ prioritizes strict budget adherence. Xin Liu shows that this mechanism can even outperform traditional projection-based methods in online convex optimization by avoiding expensive computational steps.

Applications in Stochastic Environments

This framework is particularly powerful because it is causal—it only requires the current state of the system. This makes it ideal for environments with high uncertainty:

  • Energy Harvesting: Bin Lin uses virtual queues to manage the battery states of maritime buoys, ensuring they don't deplete power before the next harvest.
  • Mobility: Tao Ouyang and Zhi Zhou apply it to "Follow Me" edge services, where the system must decide whether to migrate a service to a new node based on a long-term migration budget, without knowing where the user will move next.
  • Sustainability: Jieming Bian and Lei Wang utilize it in the CAFE framework to train federated learning models across data centers while staying within a global carbon footprint.

By converting "budgets" into "stability," Lyapunov optimization allows these diverse systems to behave intelligently in the present to protect their viability in the future.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_online_resource_management_via_lyapunov__d8e5269f,
  title = {Online Resource Management via Lyapunov Optimization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/online-resource-management-via-lyapunov-optimization/rdr_d8e5269fde164287b506e73e6dea10cb},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-22},
}

This exact version (version 1, frozen text):

@misc{lacuna_online_resource_management_via_lyapunov__d8e5269f_v1,
  title = {Online Resource Management via Lyapunov Optimization},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/online-resource-management-via-lyapunov-optimization/rdr_d8e5269fde164287b506e73e6dea10cb/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-22},
}