Real-Time Pricing Method for Spot Cloud Services with Non-Stationary Excess Capacity

Cloud operators face massive unused excess computing capacity with a stochastic non-stationary nature due to time-varying resource utilization peaks and troughs. Low-priority spot (pre-emptive) cloud services real-time pricing have been launched by many operators, which allow them maximize revenue while keeping the right reclaim capacities when scarcity occurs. However, non-stationarity of has two challenges: (1) it faces incomplete peak–trough pattern shifts in capacity, (2) suffers time space inefficiency optimal policy, needs search over large history-dependent policies state. Our objective was develop method scheme expected cumulative under We first formulated problem as Markov decision process. then developed an improved reinforcement learning algorithm obtain solution for problems. simulation experiments demonstrate that profitability proposed outperforms existing solutions. study provides both efficient optimization algorithms valuable insights into operators’ management practices.