Service or Task Duration

Service or Task duration has an important impact on the performance of distributed computations. Recall that a task corresponds to a single Service request when using Services. Make tasks long enough to compensate for communication overhead, but not so long that their interruption seriously delays the overall computation. Dividing the work into more tasks, each of which takes less time, can also mitigate the performance degradation that can arise from having tasks of different sizes. We discuss these issues in detail in the following sections.

As a running example, we use the deal valuation problem discussed in the previous section on data movement. There we assumed that each task was responsible for pricing a single deal. But this is unlikely to be efficient for most types of deals; instead, group several deals together in a single task.