This talk reports on the work Matti Hiltunen, Rick Schlichting from AT&T and myself did to identify the causes for the apparently bad performance of the Globus platform, a reference implementation of the current grid computing standards. We started with a very limited knowledge of Globus, which is a large and highly complex piece of software, and had to develop pragmatic approaches to speed up our understanding using appropriate reverse engineering techniques. In particular we used a dynamic layering technique to infer the structure of Globus and visualize the raw profiling results we obtained. This technique was instrumental in understanding how each components contributed to the system's behaviour, and what impact of the heavy use of web service technology had on Globus' performance.
This talk is a new version of the HTX'06 talk.