Figure 12 shows the average number of seconds required by plsiBackend to search for terms related to the test queries for 2, 4, 8, 12, 16, 20, and 24 processors. Figure 13 displays the average number of seconds used by plsiBackend to search for documents related to the queries using the same machine configurations. No data point is given for plsiBackend with two processors on the USENET collection since the collective memory of the two machines was not large enough to hold all the term and document vectors for the collection.
Both graphs show an overall downward trend as more processors are added to the parallel machine. However, the smaller document collections, with fewer term and document vectors, were less able to take advantage of the increased number of processors, and their curves became flatter as more processors were added. The largest document collection, the USENET collection, was able to use all 24 processors to its advantage, decreasing the time required to process queries significantly from the serial results.
