Cloud providers offer spot instances at up to 90% off compared to on-demand instances. However, spot instances have stability problems due to unpredictable interruption and difficulty in ensuring price-to-performance. In particular, this problem is in...
Cloud providers offer spot instances at up to 90% off compared to on-demand instances. However, spot instances have stability problems due to unpredictable interruption and difficulty in ensuring price-to-performance. In particular, this problem is intensified in a multi-node environment using multiple instances.
To solve this problem, this paper proposes a cost-effective and high-performance spot instance combination recommendation system Agathon. Agathon uses a multi-node spot availability dataset and quantitative hardware performance indicators to recommend an optimal spot instance combination that comprehensively considers stability, performance, and cost.
Through experiments, the necessity and effectiveness of recommendation systems specialized for multi-node environments were verified.
As a result of the experiment, the system proposed in this paper proved excellent cost efficiency by reducing the cost per task by up to 15.33% and 7.20%, respectively, compared to the price minimization strategy and performance maximization strategy. In addition, it is significant that the system is implemented as a publicly available web service to increase user accessibility.