Optimal Partitioning of Data Bases across Multiple Servers in a LAN

In 1987, Merrill Lynch was planning to establish about 450 local area networks (LANs) at its retail branch offices. Attached to the LANs would be personal computers for over 10,000 retail brokers, data-base servers providing customer and product data, and communication servers for interacting with remote systems and downloading data to the local data bases. The firm had not resolved how many data-base servers would be needed. We partitioned the data bases across multiple servers on a single LAN and developed an analytical methodology to determine the minimum number of data-base servers needed. We showed that, at most, two servers were needed for any branch office.