Aggregation methods overloading

To aggregate the responses of the SeD s, DIET uses an aggregation method which is called by the agents. This method is chosen from the SeD s by defining the aggregator type (see Section 9.3.2). By default, two aggregator types are proposed by DIET: DIET_AGG_DEFAULT and DIET_AGG_PRIORITY. Another aggregator type is also available: DIET_AGG_USER. Using this aggregator, the user can define her own aggregation method to be used by the agents. Figure 9.2 presents the global schedulers classes organization in DIET. By choosing the DIET_AGG_USER aggregator, the user commands the GlobalScheduler class to load an external module containing a UserScheduler class overloading the aggregate method.

Figure 9.2: Schedulers classes organization in DIET.

The user-defined aggregation method just needs to sort the responses from the SeD s. By locating the aggregation method on the agent, we can use different scheduling strategies which could not be implemented at the SeD level. These schedulers can also avoid some scheduling problems while submitting asynchronous jobs (with “least recently used” scheduler for example).