We ran the Andrew-Tcl benchmark to study the performance of the systems in WANs for a personal workload. Since this workload involves no data sharing, the elapsed time depends (if at all) only on the latency and capacity of the link between the client and the server. Figure 8 shows the time needed to complete the benchmark. Pangaea and Coda totally hide the network latency, because the benchmark is designed so that it reads all the source data from the local disk, and the two systems can propagate updates to other nodes in the background. On the other hand, the performance of NFS degrades severely across slow links.