Table 6 shows timings for each phase of the algorithm, omitting only time required to double check that an MPHF has been found.
In summary, our algorithm processes large key sets well, and while there is some variation in processing time because of the probabilistic nature of the operations, with an appropriate value for r the algorithm finds a MPHF with high probability.
The total time, including the Mapping step, for our algorithm to find a MPHF for the 420,878 words was 812 seconds.
With the modified implementation, a MPHF for a key set consisting of 1.2 million words was constructed on the Sequent Summetry.
As stated earlier, one application for our MPHF method is to improve access time on CD-ROMs where records addressed by single keys are sought.
Use of primary memory is greatly reduced, allowing easier MPHF constructions for very large key sets.
The experiments indicated by these rows were conducted as follows: Lower and lower bits/key values were tried until Algorithm 2 failed to find an MPHF consistently.
figure 10 illustrates that few bits/key are required, regardless of set size, but that time to find a MPHF increases rapidly as the theoritical lower bound on space is approached.
As stated at the beginning of this article, one of the most exciting aspects of this work is the wide range of applications expected for the MPHF scheme.
Our MPHF system has been suitably embedded into a dictionary manager that can support a variety of functions as can be seen in Figure 11.
MPHF indexing is performed on fields defined to be MPHF indexed.
Look up rows that have values for MPHF indexed fields that are equal to query terms.