Distributed Data Placement

A key aspect of distributed performance and functionality lies in the application of proper data placement techniques. To perform proper data placement, you should understand the manner in which each piece of data is accessed within the distributed environment. Analyzing which application or program accesses the data is not sufficient. Analyzing is merely one portion of the distributed data placement puzzle. You also need to analyze and understand the access patterns from each location on the network.

Normal data placement revolves around a single subsystem. The access patterns of programs and applications are recorded; based on that information, portions of the data are placed on DASD devices. Access-based data placement still must be done in the distributed environment. However, location access patterns must be analyzed also. Based on these patterns, portions of data can be placed at the appropriate locations within the distributed network.

The primary goal of distributed data placement is to optimize performance by reducing network transmission costs. Each piece of data should be stored at the location that accesses it most frequently. For example, storing Pittsburgh data at the Pittsburgh server makes more sense than storing it at the Chicago server. Such decisions are easy to make. Problems arise when

• A location has no server

• The frequency of access is (relatively) evenly divided between two or more servers

If the location does not have a server, place the data to the closest location on the network. For example, Pittsburgh data would be better stored in Cleveland than in Chicago, because Cleveland is physically closer to Pittsburgh than Chicago. For scenarios too close to call, the best approach is to choose a location and monitor performance. If performance is not up to par, consider migrating the data to another location.