Distributed Networks Research Portal | Auction-Based Edge Resource Allocation

Publication Overview

Why this work matters in edge computing

The paper addresses a familiar failure mode in distributed edge systems: static assignment policies leave fragmented capacity across heterogeneous nodes, causing avoidable task rejection and poor utilization under dynamic load.

Research Motivation

Low-latency edge networks must place tasks under shifting demand, heterogeneous hardware, and finite local resources. Centralized scheduling struggles in that setting.

Problem Statement

The goal is to maximize successful task placement while respecting node capacity constraints and improving fairness across the network.

Core Result

Compared with static first-fit allocation, the auction-based method reports stronger utilization, lower fragmentation, zero task rejection, and more balanced use of nodes.

Abstract

Edge computing networks need to dynamically distribute their limited and heterogeneous computation resources to handle changing tasks efficiently.

Key Contributions

Loading contributions from the paper data.

Auction Framework

Decentralized bidding for dynamic task assignment

The paper models tasks as broadcast opportunities. Feasible nodes compute bids from residual capacity and quality, and the highest eligible bidder receives the task.

Original architecture figure

Extracted from the local PDF

Auction-based resource allocation architecture figure from the paper

Bidding logic

Maximize Z = ΣΣx_ij Σx_ijC_j ≤ R_i b_pq = αA_free + β(Q/D)

Loading methodology steps.

Five heterogeneous nodes

Representative testbed spanning different capacities and hardware profiles.

Twelve incoming tasks

Task demands span small, medium, and large resource classes.

Twenty randomized trials

Average metrics are reported across repeated randomized scenarios.

Linear scaling

The bidding routine is reported with O(M · N) time complexity.

Experimental Setup

Local paper data extracted into a reusable project dataset

The portal uses task demands, reported allocations, average metrics, and figure-derived representative values taken from the local PDF. Where the paper provides only chart images, the site labels those values explicitly as digitized approximations.

Simulation environment

Platform	Ubuntu 22.04 on Intel Core i7-12700 CPU with 32 GB RAM
Trials	20 randomized runs per scenario
Network size	5 edge nodes
Task count	12 tasks per run

Task demand profile

Task	Demand	Class
Loading task data.

Data provenance note

Loading provenance notes.

Visual Analytics

Interactive plots grounded in the paper’s reported variables

Every plot below is tied to the local paper data. Distribution, capacity, and assignment views are either directly based on the tables or explicitly labeled as derived from reported figures.

DistPlot

Task demand distribution

Shows how the twelve tasks spread across small, medium, and large resource demands.

Pie Chart

Demand share by task class

Resource units are aggregated by task class to show where the workload pressure concentrates.

ViolinPlot

Per-node usage spread

Uses the figure-labeled representative run to compare how evenly each strategy loads the five nodes.

HeatMap

Task-to-node placement matrix

Visualizes which node receives each task under the two reported allocation strategies.

PairPlot

Demand vs. chosen capacity relationships

Derived from the task table, allocation mappings, and representative node capacities extracted from the paper figures.

JointPlot

Auction placements for large and small tasks

Highlights how higher-demand tasks are routed toward larger-capacity nodes in the auction-based strategy.

Reported Results

Average metrics plus representative-run views from the paper

The paper reports both averaged metrics across 20 randomized trials and specific representative-run figures. The site keeps those two views separate so the evidence trail remains clear.

Average metric comparison

Tables III and IV

Utilization vs. task volume

Figure 6 digitized values

Task allocation comparison from the paper

Figure 2. Task-to-node assignment outcome

Representative run comparison between static first-fit placement and auction-based placement.

Resource utilization by node after task allocation

Figure 3. Resource utilization per node

Figure-labeled node usage values are used in the violin and remaining-resource comparisons.

Remaining resources per node after allocation

Figure 4. Remaining resources

Shows how the auction mechanism leaves substantially less fragmented capacity across the network.

Fairness and efficiency metric comparison

Figure 5. Fairness and efficiency metrics

Summarizes lower fairness-index variance, higher utilization, more tasks placed, and fewer residual units.

Task explorer

Select a task from Table I

Select a task to compare how each strategy treats it.

Interpretation

The average metrics indicate that auction-based placement improves utilization and fairness while eliminating task rejection in the reported trials.
The representative figures show capacity consolidation rather than idle fragments spread across small nodes.
Figure 6 suggests the performance gap persists as task volume grows, which is consistent with the paper’s scalability argument.

Project Walkthrough

Video overview

The repository links the published paper to the project video so visitors can move from summary narrative to supporting material without leaving the portal.

What to expect

Problem framing for decentralized edge scheduling.
High-level overview of the auction-based assignment mechanism.
Performance outcomes relative to a static allocation baseline.
Public links to the paper, repository, and researcher profiles.

Watch on YouTube

Protected Documents

Same document security policy as your previous project

The main paper bundle now follows the same controlled access pattern used in your previous SIF-CCA project: profile verification, password request, password entry, and encrypted downloads.

Encrypted Research Assets

Protected paper and portal bundles

Visitors can still explore the public report and poster pages, but the downloadable research files are now presented through a gated workflow and password-protected encrypted archives.

Open Public HTML Report Open Poster View

Encrypted Password required

Research paper bundle

Download the IEEE paper as a password-protected encrypted archive instead of a direct public PDF link.

Encrypted Portal package

Research portal package

A second encrypted archive groups the portal files, report, poster, data, and documentation into a reusable protected package.

Public Researcher hub

Visible public materials

Your report, poster, repository, and full profile hub remain public so visitors can validate authorship before requesting protected downloads.

Open profile hub

Resources

Publication links, citation, and researcher identity

Everything here is organized for GitHub Pages deployment from the `docs/` folder and for direct use in academic portfolio or project review contexts.

Primary links

Open secure download gate HTML report Poster view IEEE Xplore entry GitHub repository

Author and profiles

Md Anisur Rahman Chowdhury

Department of Computer and Information Science, Gannon University

GitHub LinkedIn Google Scholar Portfolio ResearchGate Contact email

Citation

Loading citation.

What data in the portal is approximate?

Only the task-volume efficiency line values are approximate digitizations from Figure 6 because the local PDF does not provide that chart as a numeric table.

Why are representative figures and averages shown separately?

The paper includes average metrics across 20 randomized trials and distinct representative-run visuals. The portal preserves that separation to avoid inventing a merged dataset.

Can this be hosted directly on GitHub Pages?

Yes. The project is structured for Pages deployment from the `docs/` directory on the main branch.

Author Profiles

Highlighted researcher and public profile hub

Every major public identity channel for Md Anisur Rahman Chowdhury is collected here so authorship, collaboration, citation, and contact are easy to verify.