Introduction
As the Internet of Things (IoT) continues to expand, the demand for real-time data processing in cloud environments has surged. Efficiently scheduling a massive, unpredictable stream of tasks onto Virtual Machines (VMs) remains a significant challenge, often classified as an NP-hard optimization problem. Traditional scheduling algorithms frequently struggle with two main issues: ignoring task urgency (deadlines and value) and failing to maintain a balanced load across resources, leading to increased "makespan" and wasted capacity.
To address these challenges, researchers have proposed the Dynamic Priority and Load Balancing (DPLB) framework. This novel approach utilizes a "mutual attraction" model to map IoT tasks to cloud VMs, dynamically adjusting priorities based on urgency and value while employing real-time migration to ensure system stability.
The DPLB Framework Architecture
The DPLB framework is structured into three primary layers: the Scheduling Layer, the Virtualization Layer, and the Physical Layer. This hierarchical design allows for a modular approach to task analysis, resource mapping, and execution.
Core Innovations
The DPLB framework introduces several key mechanisms to optimize performance:
- Dynamic Priority Model: Unlike static queues, DPLB assigns priority based on Task Value Density (profit per unit size) and Urgency (proximity to deadline). To prevent "starvation," a task's priority increases the longer it waits in the queue.
- Mutual Attraction Mapping ($\rho$): The system pairs tasks with VMs using a score ($\rho$) derived from Time Attractiveness (execution speed) and Load Intensity (available capacity). Tasks are routed to the VM with the highest attraction score.
- Topic-Based Publish/Subscribe: VM "agents" publish real-time status updates to a central board, ensuring the scheduler operates on current data rather than stale metrics.
- Task Migration Manager: A background process monitors for overload events. If a VM becomes congested, tasks are proactively migrated to underutilized VMs to maintain balance.
Component Functionality
The Virtualization Layer contains several specialized modules responsible for managing the lifecycle and state of VMs, as detailed in the functional breakdown below.
Performance Evaluation
The researchers evaluated the DPLB framework using CloudSim, simulating a cluster of 20 VMs with varying processing capabilities (500–1000 MIPS) and memory. The workload consisted of 50 to 450 independent tasks with random sizes, deadlines, and user fees.
Key Results
DPLB was benchmarked against industry standards such as Min-Min, Max-Min, Genetic Algorithms (GA), and Honey Bee Behavior Load Balancing (HBB-LB).
- Reduced Makespan: DPLB achieved the lowest total completion time, with its advantages becoming more pronounced as task volume increased.
- Superior Load Balancing: The framework maintained a significantly lower standard deviation in VM load (approx. 16.70 vs. 28.25 for competitors), indicating more efficient hardware utilization.
- High Success Ratio: DPLB consistently outperformed other algorithms in meeting task deadlines.
- Increased Profitability: By prioritizing high-value tasks and optimizing execution, the framework demonstrated a clear lead in service profit per second.
Conclusion and Future Work
The DPLB framework offers a robust solution for QoS-aware resource management in IoT-cloud integrations. By treating VM availability as "attractiveness" and making task priority dynamic, it effectively balances urgency with infrastructure efficiency.
Future research will focus on extending the model to handle Directed Acyclic Graph (DAG) tasks, where inter-task dependencies exist, and incorporating security-based QoS constraints to meet the needs of sensitive IoT workflows.
Summary updated