Logarithmic Utility for Proportional Fairness

25 concepts20 papers1 proposal

In multi-user systems, resource allocation often faces a fundamental tension between two extremes: maximizing total system throughput and ensuring absolute equality. Maximizing throughput tends to favor "strong" users with high-quality connections, potentially starving "weak" users. Conversely, strict equality (max-min fairness) can significantly degrade overall efficiency by tethering the entire system to the performance of its weakest link.

Proportional Fairness (PF) offers a middle ground. It balances efficiency and equity by maximizing the sum of the logarithms of individual user utilities. This approach ensures that no single user is disproportionately deprived of resources while maintaining high aggregate performance.

The Intuition of Logarithmic Weighting

The core mechanism of proportional fairness is the logarithmic utility function, $\sum \log(U_i)$, where $U_i$ represents the utility (such as data rate or completion time) of user $i$. Mathematically, maximizing the sum of logarithms is equivalent to maximizing the product of utilities.

As Guojun Zhang et al. explain in the context of federated learning, this structure values relative improvements rather than absolute ones. A fixed percentage improvement for a participant with low utility is weighted more heavily than the same percentage improvement for a participant who is already well-served. This diminishing returns property of the logarithm naturally pushes the allocator to assist those at the bottom without completely sacrificing the gains of those at the top.

In cluster computing, Thomas Bonald and James W. Roberts argue that this approach outperforms traditional Dominant Resource Fairness (DRF). By maximizing the sum of logs of allocated shares, schedulers can ensure that jobs requiring underutilized resources are not throttled by bottlenecked ones, leading to better job completion times in dynamic workloads.

The logarithmic objective is particularly effective in environments with extreme power disparities. In full-duplex cellular networks, high-power downlink signals often interfere with and suppress low-power uplink signals. Itsikiantsoa Randrianantenaina et al. demonstrate that optimizing for aggregate throughput in these systems would effectively silence the uplink. By employing a sum-log-rate utility function, the system enforces fairness, preventing the spectral starvation of mobile users and maintaining connectivity for vulnerable links.

Similar principles apply to broadcasting and small-cell management:

  • Broadcasting: Aria HasanzadeZonuzy et al. show that incorporating logarithmic utility into broadcast protocols prioritizes nodes that have received fewer packets, preventing high-traffic areas from starving remote or poorly connected nodes.
  • Small Cells: Ursula Challita and Mahesh K. Marina use a log-sum utility to balance traffic across licensed and unlicensed bands, ensuring that LTE small cells coexist fairly with WiFi users.
  • Optical Networks: In Passive Optical Networks (PONs), Genya Ishigaki et al. use logarithmic functions to ensure that even nodes with low traffic loads receive a minimum guaranteed resource window, maintaining system-wide stability.

Decentralization and Quality of Experience

A significant advantage of the proportional fairness framework is its suitability for decentralized implementation. Because the logarithmic objective is concave, it often allows for elegant, distributed solutions.

For instance, Xiong Wang et al. utilize logarithmic rewards in a Multi-Armed Bandit framework for edge computing, allowing multiple users to share server resources autonomously without a central controller. In the energy sector, Phuthipong Bovornkeeratiroj et al. apply logarithmic utility to individual solar node outputs. This forces nodes to reduce their output proportionally to their capacity during grid congestion, achieving a Pareto optimal state where no participant can increase their share without harming others.

Furthermore, the framework can be adapted to prioritize human-centric metrics. Ping Wang et al. propose a scheduling scheme that uses a logarithmic function of predicted Mean Opinion Scores (MOS). By assigning higher priority to users with lower predicted satisfaction, the system balances raw technical throughput with actual Quality of Experience (QoE), preventing resource starvation for users in poor signal conditions.

Summary of the Mechanism

Proportional fairness works because the logarithmic transformation changes the optimization goal from "maximize the total" to "maximize the balance of gains." Whether applied to pilot reuse in Massive MIMO systems—where Jy-yong Sohn et al. found that optimal pilot assignment scales logarithmically with priority weights—or to wireless power transfer, the result is a system that is both efficient and robust against the starvation of its individual members.

Key Foundational Work

Max-Min Fairness in 802.11 Mesh Networks (2010) — contributed 2 concept elements to this direction.
On Cross-Layer Design for Streaming Video Delivery in Multiuser Wireless Environments (2006) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_proportional_fairness_utility_46ec9f86,
  title = {proportional fairness utility},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/proportional-fairness-utility/rdr_46ec9f863107444799a20a627ce5834c},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-21},
}

This exact version (version 1, frozen text):

@misc{lacuna_proportional_fairness_utility_46ec9f86_v1,
  title = {proportional fairness utility},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/proportional-fairness-utility/rdr_46ec9f863107444799a20a627ce5834c/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-21},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake