On-line modification of long-lived distributed programs

A number of software systems are now being designed with the requirement that the system have both high availability and an extended run-time that will be measured in terms of years. A problem then occurs when these systems must be modified, whether for correcting existing problems in the system or for adding features to the system. The traditional methodology for making modifications to a software system has been to produce a modified version of the software, take the executing system off-fine, replace the old version of the software with the modified version, and bring the system back on-line. This cycle of taking the system off-line and later bringing it back on-line conflicts with the design goal of high availability. To this end, it would be desirable to construct these systems in such a way that they could be modified while the system remains on-line.; This work addresses the problem of on-line system modification in the context of a distributed Ada 95 program. First, a taxonomy is presented that can be used to break the problem of on-line modification down into a number of subcases. Next, a methodology is presented that can be used to build systems that fall into two of these subcases. This methodology is split between low-level protocols that are provided for the user as part of a special support package, and high-level protocols and design patterns that the user must use to produce systems that allow on-line modification.; A proof-of-concept implementation has been constructed to allow the implementation of several demonstration example programs.