| Usually, computing-intensive application requires scalable computing resources.Traditional high performance computing platform provides large scale computing resources, but users need to buy such computing infrastructures. With the emergence of cloud computing, more and more scientists and engineers migrate their application from local computer to the cloud. However, researchers have to spend much time on programming and spend a lot of money on migration.This paper presents a lightweight cloud-based parallel distributed computing platform to support compute-intensive and data-intensive scientific computing namely Simulation Runner. With the help of Simulation Runner, scientists simply submit their embarrassingly parallel compute-intensive or data-intensive applications with appropriate arguments without cloud development experience or purchasing and maintaining computer infrastructures. Scientist no longer needs not to modify their embarrassingly parallel scientific application. The system divides the job into tasks, so as to run their tasks parallel at instance level. When the task running on the computer instance, the system record their standard output, standard error. When the task finished, the system upload the result file back in to the Azure Blob Storage. In order to increase the efficiency of the system, the system also uses file cache management to improve performance and reduce cost for data-intensive computing tasks. The system also support multiple data source to upload the job package, the system also allow users to upload their package from Dropbox, OneDrive and Azure Blob Storage. In several situations,users’ scientific applications not only depend on the windows runtime environment, but also need other runtime environment, like Python, MATLAB, R. Test results for user experiences have been with Simulation Runner positive. A comparison study on the performance and cost between Simulation Runner and on premise machines is given at the end of the paper. |
PDF Full Text Request