Synopsis

• Why do we want distribution? Let's remember these goals: no use cheating.
• In order to support scalability, we must provide late distribution: allow to reconfigure the distribution without invalidating client software.
• A first step is to provide location transparency, but still, there are more serious reasons why traditional sequential software is not distributable
• The next step is to set for an event driven architecture, and to provide availability through using threads. The shortcomings of this solution lead to the next increment.
• With transactions, we are pretty close, but not quite on the track.
• The concept we are missing is this of scope, and we are missing it because its traditional acception has been badly modified by the introduction of time into the picture. Scopes must be identified so as to be kept up-to-date (reactivity).
• What we need is an explicit support for temporal scopes. This is what we call operations. These are obviously classes (prototyped scopes). Their implementation involves genericity, i.e. templates.
• In what do they fill our requirements and solve our problems?
• Conclusion
• Related work

Table of contents