Pilot
The following is a scenario of how UCR is used in SRM 4.5, a production intranet site that provides reports on systems performance via the Web.
IGS uses two diverse performance data sources:
- SRM (Server Resource Management) – measures daily resources of servers such as CPU utilization, percent of memory used, number of users logged in, etc.
- EPP (End-to-End Probe Performance) – measures the response time of probes such as round-trip email messages, access to specific web pages, etc.
IGS would like to create reports that contain information from both of these sources. Ideally, a report should allow examination of EPP data, along with details from SRM data, in order to analyze what caused poor performance of a probe. For example, when a probe fails and indicates problematic performance, it is desirable to analyze which server resources are at the root of the problem. Similarly, server resources can be intentionally overloaded to see how problem responses will be effected under these conditions. These details can be analyzed using the UCR reports.
Similarly, IGS is interested in investigating SRM data while using the EPP data to analyze how a probe performs at a specific load on a server. For example, reports will generate summaries of server performance per day. By using both server and probe data, you can check which probes caused the server to overload.
By looking at the different data sources and seeing how the information is related, IGS gets a better breakdown of the entire situation.
Current techniques generate a single report that unifies these data sources by creating a new data warehouse and copying the data from the original sources to the new warehouse. In contrast, UCR does not copy data from the sources to a new unified warehouse. Using UCR, mappings are defined from the data sources to a unified schema represented in XML. Next, a Lookup Engine unifies the data using those defined mappings. All queries are then addressed against the unified schema. The resulting report shown below is visualized using HTML and Java applets. The applet is based on the LifeLines component taken from the Visual Analysis group at IBM Watson.
Click to see full size
