Efficient browsing and retrieval of geographically referenced information requires the allocation of data on different storage devices for concurrent retrieval. Earlier work on the data placement problem using wavelet decomposed image data focused on the evaluation of various placement strategies using a disk simulator. Efforts were made to replace this simulated portion of the evaluation system with real magnetic disks. To this end, dedicated partitions were acquired on four separate disks for the purposes of data placement independent of the regular Unix file systems. In order to control the placement of data in these partitions, the partitions are processed as "raw" devices through the Unix file I/O system without the usual buffering. This approach eliminates the need to write complex and time consuming disk drivers.
Initially experiments were conducted to measure the disk parameters before embarking on the actual data placement. However, due to the fact that the disks are also being used for the regular file systems, it was not possible to obtain such results. The reason is that direct control of the disks at the SCSI level cannot be obtained unless the complete disk is dedicated for our purposes - the use of part of the disks for the regular file system makes operation at the SCSI level difficult and risky (possible loss of file system data) since there is need to disable on-disk caching etc.
Consequently, the task of measuring the disk parameters was postponed until dedicated disk drives are available. Instead, the real disks were integrated into the evaluation system so that the performance could be measured, even though these figures may be affected by the use of the disks for the file system. The results of these experiments have collaborated the findings from the simulations. The new results were reported in an updated version of the earlier paper [6].
In a more general setting of geographically referenced information that is not necessarily wavelet decomposed, we approach the data placement problem by dividing a two dimensional space into tiles. The system can allow users to specify regions of interest using a query rectangle and then retrieving all information related to tiles overlapping with the query. Necessary and sufficient conditions were derived for strictly optimal allocations of two-dimensional data. These methods, when they exist, guarantee that for any query, the minimum number of tiles are assigned the same storage device, and hence ensures maximal retrieval concurrency [8].