Distributed join processing performance evaluation

Efficient join acceleration methods that combine hybrid hashing, join indices and domain vectors with semi-join techniques to accelerate distributed joins are developed. An analytic performance model which allows for parallel execution of the methods at each join site is developed to compare the total task duration time of the methods. The methods are compared over a wide range of memory sizes, semijoin selectivities and network transmission speeds. The distributed join index method performs well if memory size is large and selectivity is low. The distributed hybrid hash method always has the advantage that it requires no residual data to be stored between runs, however performance is poor at low network transmission rates due to high pre-semijoin transmission requirements. The distributed domain vector hash method performs well in high selectivity conditions.<
>