4. Conclusions and Discussions
In this study, we analyzed three approaches to Query Preview Applications with multi-valued attributes. First method, which is already in use by the current Query Preview Panels, requires the transfer of the data set over the network and then runs local queries on the client. Second method pre-calculates the finite number of results for the possible queries to be asked on a Query Preview Panel and sends answers to the client upon request (post-fetch method). The final method (pre-fetch method) transfers all the pre-calculated answers before the querying session, to the client site, where with this approach the designers of the system need to bring some restrictions to the Query Preview Panel to reduce the size of the pre-calculated answers.
For small databases and especially with high network bandwidth, the simple data set transfer scheme might be used without any trouble. This might lead to better memory utilization and save database administratorsí time as they do not need to create the pre-calculated answers regularly. On the other hand, with slow networks or with highly congested links, transferring the data over the network might be problematic. More over only a few data collections of NASA qualify for this approach as NASA data sets are generally much bigger than the commonly used data sets of the business applications.
The pre-calculation based schemes although requires regular updates might solve some of the problems of the current applications. Most of the attributes of the current Query Preview Panels can either be restricted to have less number of possible selections or the granularity of the attributes used with these attributes can allow database administrators to create reasonably sized files for the pre-calculated answers.
Still, if the network bandwidth is low or the size of the pre-calculated answers is bigger than the data set size we recommend the usage of the post-fetch approach instead of the pre-fetch approach. In most cases, the data set size will be much bigger than the pre-calculated set especially if there are restriction on the Query Preview Panel.
One possible disadvantage of the post-fetch method is the requirement of continuous communication between the client and the server. Although we propose that the connection will be required only upon request, lengthy querying sessions might suffer from connection time-outs.
All of these approaches were proposed primarily for the Internet and the World Wide Web users. Hence, we assume that many of these users have the capability of connecting to the NASA servers via some network browsers such as Netscape or the Internet Explorer. Users with gopher and basic telnet capabilities will not benefit from these approaches as much as the others.
As future work, we feel the need for a more detailed testing and experimentation of the approaches. Tests with real users and varied network work-load and bandwidths should be performed. Also, new approaches for users with only gopher and telnet capabilities should be developed.