Cloud Scaling with Narrative-Driven Computing - An approach to designing and implementing algorithms with intrinsic explainability

Abstract

With the emergence of cloud computing, allowing for flexible and scalable infrastructures, dynamic resource allocation has become an important aspect of service management. An optimized resource allocation strategy can improve both cost- and QoS-related aspects; thus, the challenge has received significant attention in academia, and several sophisticated methods proving exceptional performance have been proposed. However, despite their superiority, there is seemingly a low adoption of these methods in the industry.

This thesis attempts to approach the disparity between academia and the industry by proposing a novel method to design and develop models centered around increased explainability to foster trust and facilitate industry adoption. Through the process of further designing and developing an existing scaling algorithm according to our initially proposed "narrative method," which embraced the use of metaphors, narrative-driven computing is presented as the final proposal based on the experiences we had along the way. Narrative-driven computing is a new approach to designing and developing models where meaningful functionality is wrapped into metaphors that constitute the necessary building blocks for "explainability."