Publications

The desire to provide a measure of computer network availability at Sandia National Laboratories has existed for along time. Several attempts were made to build this measure by accurately recording network failures, identifying the type of network element involved, the root cause of the problem, and the time to repair the fault. Recognizing the limitations of available methods, it became obvious that another approach of determining network availability had to be defined. The chosen concept involved the periodic sampling of network services and applications from various network locations. A measure of ''network'' availability was then calculated based on the ratio of polling success to failure. The effort required to gather the information and produce a useful metric is not prohibitive and the information gained has verified long held feelings regarding network performance with real data.

As information systems have become distributed over many computers within the enterprise, managing those applications has become increasingly important. This is an emerging area of work, recognized as such by many large organizations as well as many start-up companies. In this report, we present a summary of the move to distributed applications, some of the problems that came along for the ride, and some specific examples of the tools and techniques we have used to analyze distributed applications and gain some insight into the mechanics and politics of distributed computing.