Section: New Results

Experimentation and Visualization in Large Systems

Despite a widespread belief regarding the simulation of large-scale computing systems, we showed in [15] that achieving high scalability does not necessarily require to resort to overly simple models and ignore important phenomena. In fact, by relying on a modular and hierarchical platform representation while taking advantage of regularity when possible, we were able to model systems such as data and computing centers, peer-to-peer networks, grids, or clouds in a scalable way. Finally, in [34] , we examined the ability to conduct consistent, controlled, and repeatable large-scale experiments in areas of computer science where availability, repeatability, and open sharing of electronic products are still difficult to achieve.

We also discussed in [22] the concept of the reconstructability of software environments and we proposed a tool for dealing with this problem. In a similar vein, we developed Expo [41] , a tool for conducting experiments on distributed platforms. Our experiments confirmed that Expo is a promising tool to help the user with two primary concerns: how to perform a large scale experiment efficiently and easily, together with its reproducilibity.

The exponential number of processes that are executed in high performance applications and the very detailed behavior that we can record about them lead to a trace size explosion both in space and time dimensions. Thus, if the amount of data is not properly treated for visualization, the analysis may give the wrong idea about the behavior registered in the traces. We dealt with this issue in [38] in two ways: first, by detailing data aggregation techniques that are fully configurable by the user to control the level of details in both space and time dimensions, and second, by presenting two visualization techniques that take advantage of the aggregated data to scale.

Furthermore, given that the performance of parallel and distributed applications is highly dependent on the characteristics of the execution environment, the network topology and characteristics directly impact data locality and movements as well as contention. Unfortunately few visualization available to the analyst are capable of accounting for such phenomena, so we proposed in [39] an interactive topology-based visualization technique based on data aggregation that enables to correlate network characteristics, such as bandwidth and topology, with application performance traces. Such visualization techniques enable us to explore and understand non-trivial behaviors that are impossible to grasp otherwise and the combination of multi-scale aggregation and dynamic graph layout allows us to scale the visualization seamlessly to large distributed systems.