Questions/Comments

version 1 (see Old Versions)

  • This diagram is obtained by zooming into the Feedback actor on the Feedback Control System.
  • It represents variations of patterns in style "separate definition of pattern combinations".
    • mw: Interesting, we should play with this some more. At first, I felt overload, and had questions about scalability, if you write something like (ST-OB-BB). I must say I felt that Chung's style which looked at all the NFRs and the patterns, independent from the actors, in one diagram clearer. In his diagrams, each pattern had a separate node, and a combination of patterns would be represented by satisfied/denied labels. But this is really an area, where some improvements need to be made in GRL: how do we best represent that a group of operationalizations together constitutes an alternative.
  • It does not show the dependency links.
  • the second diagram could be extended to show all pattern combinations (not just the ones considered for Feedback) * mw: Two comments: (1) you want to limit what you show to what is relevant to a view, but (2) you want to see the impact of the patterns on the system as a whole (eg by looking at each actor separately, we cannot determine that maybe using a particular pattern has benefits from multiple actors/goals, not just the current one).
  • mw: Persistence contributes to accuracy (ie better control strategies), but could (really depends, though) also be required for audit purposes (going on a limb here, because you would not do that for a regular control system)
  • mw: Blackboard (Shaw) and Active Repository (Zdun) are really the same thing, simplifying our diagrams a bit. There is also Blackboard (Buschmann) -- you guessed it's popular! -- which is the AI architecture with knowledge sources/agents sharing information through a coordination medium, which is also found in (reactive) tuple spaces.
  • mw: There should be a complexity or some goal like it (maintainability?) to indicate that Observer is more complex than Producer-Consumer (which Buschmann classifies as a variant of Observer for relationships that are 1:1; in the pattern structure this simply means that we leave out some components/connections, eg no loop to notify multiple parties, an no attach/detach events). On the other hand PC can easily grow into OB later on.

version 2 (see Old Versions)

version 3 (see Old Versions)

  • typos

version 6 & 7 (see below)

  • v6 shows the choice made by the book (POAD)
  • v7 shows alternative based on strategy and active repository

GRL Diagrams

Alternative

-- Gunter Mussbacher - 22 Mar 2006

Topic revision: r6 - 26 Mar 2006 - 23:47:18 - Gunter Mussbacher
Work.GoalFeedback moved from Work.ActorFeedback on 23 Mar 2006 - 22:53 by Gunter Mussbacher - put it back
 
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback