CoRI: Collectors of Ressource Information

CoRI manages access to different tools for collecting information about the SeD. Currently, various tools, called collectors, are implemented: CoRI Easy and CoRI batch. The user can choose which collector will provide the information.

The idea is that there are various levels of expertise among users developing SeD code and DIET client code:

1. Basic usage will rely on default scheduling strategies.
2. More advanced users will perhaps want to tweak the basic scheduling strategies, by providing different priorities between performance metrics (DIET_AGG_PRIORITY), but still using the default performance metrics provided by DIET.
3. Even more advanced users can:
   • ask CoRI to fill the estimation vector with particular sets of metrics,
   • or explicitely add to the default metrics, or replace the default metrics, with their own custom metrics in the estimation vector;
   • and optionnaly implement their own custom plugin scheduler (DIET_AGG_USER), especially if opaque (not floating point, so unsortable with DIET_AGG_PRIORITY style scheduling) metrics are set in the estimation vector.

To allow all these use cases, CoRI, by default, fills the estimation vector with some default performance metrics. CoRI is also designed to allow managing high level CoRI performance metrics collectors (see figure 10.1), which will on demand fill the estimation vector with a set of predefined performance metrics. Finally, as seen in chapter 9, it is possible to manually fill the estimation vector with any custom metrics.

Figure 10.1: CoRI overview