A Debt-Aware Learning Approach for Resource Adaptations in Cloud Elasticity Management
This addresses the trade-off between economics and performance in autonomous cloud resource management, though it appears incremental as it builds on existing reinforcement learning methods.
The paper tackles the challenge of mismatched resource demand and provision in cloud elasticity management by proposing a debt-aware reinforcement learning approach, which achieves higher utility for cloud customers while maintaining performance levels.
Elasticity is a cloud property that enables applications and its execution systems to dynamically acquire and release shared computational resources on demand. Moreover, it unfolds the advantage of economies of scale in the cloud through a drop in the average costs of these shared resources. However, it is still an open challenge to achieve a perfect match between resource demand and provision in autonomous elasticity management. Resource adaptation decisions essentially involve a trade-off between economics and performance, which produces a gap between the ideal and actual resource provisioning. This gap, if not properly managed, can negatively impact the aggregate utility of a cloud customer in the long run. To address this limitation, we propose a technical debt-aware learning approach for autonomous elasticity management based on a reinforcement learning of elasticity debts in resource provisioning; the adaptation pursues strategic decisions that trades off economics against performance. We extend CloudSim and Burlap to evaluate our approach. The evaluation shows that a reinforcement learning of technical debts in elasticity obtains a higher utility for a cloud customer, while conforming expected levels of performance.