The simulator allows us to control the number of clients and servers, their connectivity, and also workload issues such as how the database is distributed among the servers, how clients share, and what transactions are like. The details of the algorithm are simulated precisely; we replaced client-server tuples in multistamps with server stamps when there were at least 10 entries for the same server. We constructed the simulator and workloads starting from earlier concurrency control studies [6][13]. These studies were performed for a single-server, multi-client system; we extended them to a distributed database with multiple servers. We ran experiments with varying values of simulator parameters; we present results for a particular setup and mention results of other experiments along the way. Figure 2 describes the parameters of the study.
Figure 2: Summary of Parameter Settings