Although the timings were taken on dedicated machines, the network was necessarily shared by many users when the timings were taken. In particular, since the timings were taken during the early morning hours, each search engine had to compete with both incremental and full backups of disks occurring over the network, as well as a daily software distribution to each machine. Even though the network shared by the dedicated machines was relatively isolated, the term and document vectors for each document collection were stored on disks outside the cluster of dedicated machines. LsiQuery, lsiBackend, and lsiFinder all load the term and document vectors from disk into memory as they are needed, making all three of them dependent on the speed of the network. Since little could be done to prevent other users from accessing the network while the timings were being taken, the network was monitored during testing to ensure that the network stayed at approximately the same load while each search engine was running. However, since accurately monitoring the network was difficult, the times reflect natural variances as the network became more and less congested.