An Autoscaling System Based on Predicting the Demand for Resources and Responding to Failure in Forecasting.

Area of Science:

• Computer Science
• Cloud Computing
• Edge Computing

Background:

• Edge computing and sensor technologies are revolutionizing real-time data processing.
• Data acquisition involves collecting sensory information (images, videos) and transmitting it to the cloud for analysis.
• Proactive resource allocation is crucial for handling varying data volumes and request frequencies in real-time.

Purpose of the Study:

• To propose a framework for dynamic resource provisioning in cloud infrastructure for real-time data processing.
• To address the risks of system failure associated with solely predictive resource allocation.
• To enhance the performance and efficiency of autoscaling algorithms in edge computing environments.

Main Methods:

• Developed a framework with algorithms for periodic monitoring of resource requirements.
• Implemented dynamic adjustment of resource provisioning to match actual demand.
• Conducted experiments using the Bitbrains dataset with specific network throughput and threshold settings.

Main Results:

• The proposed system achieved a 99% performance improvement in autoscaling.
• The system introduced only 0.43 ms of additional computational overhead compared to prediction models.
• Demonstrated effective resource management for real-time data processing under experimental conditions.

Conclusions:

• Dynamic resource provisioning offers a more robust solution than purely predictive models for real-time data processing.
• The proposed framework enhances autoscaling performance and efficiency in edge computing.
• This approach mitigates risks of system failure by adapting to actual resource demands.