Accelerating Data-Intensive Applications: a Cloud Computing Approach to Parallel Image Pattern Recognition Tasks

摘要

Performance is an open issue in data intensive applications, such as image pattern recognition tasks. To process large-scale datasets with high performance more resources and reliable infrastructures are required for spreading the data and running the applications across multiple machines in parallel. The current use of parallelism in high performance computing and with multicore hardware support is costly and time consuming. To remove the burden of building, operating and maintaining expensive physical resources and infrastructures, Cloud computing is emerging as a cost-effective solution to address the increased demand for distributed data, computing resources and services. In this paper, we explore and evaluate parallel processing performance of an image pattern recognition task in the Life Sciences based on a Cloud computing model: Infrastructure-as-a-Service. Namely, we rent computing infrastructures from cloud providers. We have developed the image pattern recognition task in both sequential and parallel ways, deployed them, and conducted our experiments on cloud infrastructure. The performance has been evaluated using speedup as a measurement. We have calculated the cost of our experiments, which demonstrates that cloud computing could be a cheaper alternative to supercomputers and clusters given this task.