Containers Orchestration with Cost-Efficient Autoscaling in Cloud Computing Environments

Containers are standalone, self-contained units that package software and its dependencies together. They offer lightweight performance isolation, fast and flexible deployment, and fine-grained resource sharing. They have gained popularity in better application management and deployment in recent years and are being widely used by organizations to deploy their increasingly diverse workloads such as web services, big data, and IoT in either proprietary clusters or cloud data centres. This has led to the emergence of container orchestration platforms, which are designed to manage the deployment of containerized applications in large-scale clusters. The majority of these platforms are tailored to optimize the scheduling of containers on a fixed-sized private cluster but are not enabled to autoscale the size of the cluster nor to consider features specific to public cloud environments. In this work, we propose a comprehensive container resource management approach that has three different objectives. The first one is to optimize the initial placement of containers by efficiently scheduling them on existing resources. The second one is to autoscale the number of resources at runtime based on the current cluster’s workload. The third one is a rescheduling mechanism to further support the efficient use of resources by consolidating applications into fewer VMs when possible. Our algorithms are implemented as a plugin-scheduler for Kubernetes platform. We evaluated our framework and the effectiveness of the proposed algorithms on an Australian national cloud infrastructure. Our experiments demonstrate that considerable cost savings can be achieved by dynamically managing the cluster size and placement of applications. We find that our proposed approaches are capable of reducing the cost by 58% when compared to the default Kubernetes scheduler.

💡 Research Summary

The paper presents a comprehensive resource‑management framework for containerized applications running on public‑cloud infrastructures, implemented as a Kubernetes plugin scheduler. Recognizing that most existing container orchestration platforms (Kubernetes, Docker Swarm, Apache Mesos, etc.) are designed for fixed‑size private clusters and lack integrated autoscaling and rescheduling capabilities, the authors propose three tightly coupled algorithms: (1) an initial placement algorithm that minimizes the number of worker VMs while satisfying CPU and memory requests of both long‑running services and batch jobs; (2) a runtime autoscaling mechanism that monitors cluster‑wide resource utilization and dynamically provisions or de‑provisions VMs based on configurable high‑ and low‑watermarks, taking into account the current container placement and the movability of workloads; and (3) a rescheduling procedure that, when a new task cannot be scheduled, attempts to rearrange movable containers across nodes to free up resources, using a two‑phase approach (lightweight intra‑node consolidation followed by heavyweight cluster‑wide rebalancing). The rescheduling logic leverages Kubernetes features such as PodDisruptionBudget to ensure zero‑downtime migrations for stateless or replicated services.

The framework is evaluated on the Australian National Cloud platform with heterogeneous workloads (CPU‑intensive, memory‑intensive, and mixed). Experiments demonstrate that the initial placement alone yields roughly a 20 % reduction in operational cost compared with the default Kubernetes scheduler. When combined with the autoscaling and rescheduling components, total cost savings reach 58 %, while latency and SLA violation rates improve significantly. The study also shows that the integrated approach reduces the number of active VMs, shortens container launch waiting times, and maintains service performance during scale‑in events.

Compared with related works, which typically focus on a single aspect (e.g., initial placement using bin‑packing, ant‑colony optimization, or genetic algorithms) and often rely on simulations, this work delivers a full‑stack solution validated on real cloud infrastructure. It uniquely incorporates cloud‑specific considerations such as per‑billing‑period pricing, unlimited VM provisioning, and the need to balance cost against QoS. The authors discuss limitations (homogeneous VM types, reliance on movability annotations) and outline future research directions, including support for heterogeneous VM families, predictive workload forecasting, and multi‑objective optimization that also accounts for energy consumption. Overall, the paper makes a strong case for tightly coupling container scheduling with cloud‑native autoscaling and rescheduling to achieve substantial cost efficiency without sacrificing performance.