Next: Uniform Strategies Up: T. Zurek: Optimisation of Previous: Conclusions

Partitioning Strategies

In this chapter, we present several families of partitioning strategies, all of which can be based on the information stored in an IP-table. These strategies create partitions for computing a symmetrically partitioned temporal join (3.6):

of two relations R and Q. There are many goals according to which the fragments R₁, ..., R_m, Q₁, ..., Q_m can be created. For example, one could aim to minimise the processing costs. However, this task is not that easy due to the complexity of the performance model (see chapter 8). Even relatively simple constraints, such as the ones for IP (see chapter 5), can necessitate a very expensive calculation in order to find a suitable partition. This leads us to consider alternative goals, such as the efficiency of the partitioning strategy itself.

In the following, several goals, and the family of strategies that result from it will be discussed. We thereby concentrate on the most general goals and strategies. All the algorithms that are used in that context van be efficiently implemented using IP-tables. In the remainder, we adopt the notation of complete IP-tables. Nevertheless, all of the techniques and algorithms that are described can be used in conjunction with incomplete IP-tables too, possibly at the expense of a decreased quality of the result. If this is the case we will point to this fact.

Next: Uniform Strategies Up: T. Zurek: Optimisation of Previous: Conclusions

Thomas Zurek