Spot Price Prediction Based Dynamic Resource Scheduling for Web Applications

Resource scheduling algorithms are crucial for cloud web applications to minimize resource rental costs of virtual machines elastically rented from public clouds while guaranteeing performances. Most existing algorithms only focus on virtual machines with fixed prices without considering spot virtual machines which have dynamic and cheaper prices. A few existing algorithms have considered the renting of spot resources to decrease rental costs. However, trends of dynamic prices are not considered which are beneficial to decrease rental costs further. Meanwhile, although spot virtual machines are cheaper, fluctuated prices are likely to produce out-of-bid events which make spot virtual machines unreliable. In this paper, a spot price prediction and minimum renting interval based resource scheduling method is proposed. Spot prices are predicted to identify cheap and stable spot types to improve the reliability. Frequent spot type switching is avoided in some extent by setting a minimum renting interval. The proposals are evaluated on CloudSim and experimental results show a better performance than existing algorithms.