Abstract

This paper explores the design and implementation of a cloud-based burst buffer system aimed at accelerating I/O performance for data-intensive applications in cloud environments. By leveraging the high throughput between compute nodes, the system buffers application data, effectively mitigating the I/O bottlenecks typical in shared cloud storage. A prototype implementation on Amazon EC2/S3 demonstrated significant performance enhancements, achieving up to 2.9 times improvement in execution time for real-world applications and reducing monetary costs by 50%. The proposed system not only addresses the critical issue of low I/O throughput but also offers scalable and cost-effective solutions for high-performance computing in the cloud.