We believe that, along with communication networks and distributed database systems, parallel computation will be one of the main supporting technologies for future digital library systems. The tremendous rate of development in hardware/software technology means that any conclusions based on present capabilities will be outdated during the life cycle of any short term project. Therefore real computational bottlenecks need to be identified. For spatial digital libraries, there are a number of aspects from which serious computational bottlenecks arise. For example (1) the enormous, and rapidly growing, quantity of raw digital datasets, (2) the demand for increased resolution and size of images both for browsing and for scientific applications, (3) the distributed nature of both data and potential users, (4) the increase in the sophistication of queries.
The future of spatial digital libraries depends in part on the availability of parallel means to overcome such computational problems. Our aim is to develop the necessary theory and computational tools to this end, as well as to validate our developments and algorithms on real-world data. When computations on very large data sets are involved, load balancing, routing, data partitioning, locality, and prefetching become important implementation parameters that can drastically improve the system's response time [23]. These issues exist in a distributed system of clients/servers, especially where one or more of the servers is itself a multiprocessor. However it is virtually impossible to create a meaningful application if a system designer is confronted with all the complexities of dynamic load balancing, data migration, and their resulting communication patterns. Functional designers need to be shielded from the complexities of the underlying architecture to the greatest extent possible, at the same time being able to develop applications that get reasonable performance. We see the necessity of parallelization in this area as analogous to the situation that existed prior to the widespread use of virtual memory in sequential systems.