A Web-based distribution protocol for large scale analysis and optimization applications

Zhou, Qun, Mahdavi, Babak, Liu, Dao Jun, Guibault, Francois, Ozell, Benoit and Trépanier, Jean-Yves (2003) A Web-based distribution protocol for large scale analysis and optimization applications. Kent, R. and Sands, T. (eds.) In High Performance Computing Systems and Applications. Springer. pp. 257-270 . (doi:10.1007/978-1-4615-0288-3_29).

Abstract

This paper presents a distribution and remote execution protocol for computationally intensive analysis applications. This protocol, developed in the context of the VADOR project, aims to encapsulate data transfers and remote execution of commercial and in-house applications used for aeronautic design optimization at Bombardier Aerospace. While a companion paper describes the overall framework architecture of the VADOR project, this paper concentrates on distribution aspects of applications involved in the analysis processes. Execution of analysis processes are mediated through a framework-specific execution server, that launches execution through the invocation of services provided by distributed CPU server machines. The CPU servers take care both of the data transfers involved in each invocation, and of the execution of the application. Data transfers are performed using standard tools and web-based protocol. Application execution and control, on the other hand, is taken in charge by custom “wrapper servers” running on each CPU. This paper discusses design considerations of the “wrapper servers”, which include the flexible and adaptable wrapping of legacy and commercial applications in use at Bombardier. The paper centers around communication issues between the execution and wrapper servers, and data transfer issues, which involve peer transfers servers at each end of the communication stream. Error handling and scalability will be specifically addressed in the context of the large-scale computations involved by the analysis processes. Security considerations will be discussed as well.

This record has no associated files available for download.

Contributors

Download statistics