Advance Reservation Scheduling in Cloud Computing: A Review of Models, Algorithms and Practical Challenges

Abstract

Advance reservation (AR) extends conventional cloud resource management by allowing consumers to request compute capacity for a specified future time window with explicit start and end constraints. This capability is increasingly relevant for deadline-driven analytics, production batch workloads, time-bound digital services, and regulated environments that require predictable capacity rather than best-effort elasticity. However, AR scheduling in cloud systems remains challenging because providers must balance temporal commitments against uncertain demand, heterogeneous infrastructure, energy and thermal constraints, admission control risk, and multi-tenant fairness. This review synthesizes recent research on AR-oriented scheduling and closely related “time-constrained” provisioning approaches, including scheduled virtual machine demands, reservation-aware placement, and capacity planning under uncertainty. We organize the literature around core modeling assumptions, algorithmic families, and evaluation practices, and we connect these to cloud-specific implementation realities such as virtualization overhead, migration feasibility, and policy enforcement. The review highlights progress in meta-heuristic placement for time-window requests, multi-cloud allocation under reserved and on-demand pricing, learning-guided capacity planning with short- and long-term reservation decisions, and scheduling strategies that integrate operational sustainability objectives. Finally, we identify persistent gaps—particularly in realistic benchmarking, provider-grade admission control, and cross-layer coordination—and outline practical research directions for robust, verifiable AR scheduling at scale. Strengthening evaluation realism, integrating uncertainty and risk, and aligning scheduling designs with provider implementation limits are the most important next steps for enabling dependable AR at cloud scale.