An online service provisioning strategy for container-based cloud brokers

Abstract

Container-based cloud brokers sit between multiple cloud providers and users, renting virtual machines (VMs) from multiple cloud providers, provisioning services to users, and executing jobs in the form of containers. Current cloud providers offer two billing models for Infrastructure-as-a-Service (IaaS), including on-demand and reserved. Given the huge uncertainty of user requests, cloud brokers can combine two billing options to rent VMs to reduce costs while gaining the flexibility to handle short-term fluctuations in requests. Optimal scheduling and auto-scaling strategies depend on information about user requests over a long period of time in the future, which is often difficult to predict accurately. To solve this problem, this work proposes an online service provisioning strategy to optimize the cost of cloud brokers to satisfy user requests without requiring any future information. Our proposed strategy consists of an online container scheduling algorithm and an online reserved VM pool auto-scaling algorithm. It is shown theoretically that the two proposed online algorithms can achieve guaranteed competitive ratios of n+1n−1TT−3(μ+1) and 2−1T respectively. Eventually, the effectiveness of our proposed algorithms is validated with both simulated and multiple real user request datasets. Our proposed strategy can significantly reduce the cost of cloud brokers compared to several benchmarks.