MINDMAKERS Forum
Welcome, Guest. Please login or register.
September 08, 2010, 02:00:55 PM

Login with username, password and session length
Search:     Advanced search
NB: Spam bots are becoming smarter every day - we had to turn off regular registration. To become member, please send email to Kris Thorisson ([kris'_lastname] att ru dott is).
337 Posts in 99 Topics by 99 Members
Latest Member: peterwit
* Home Help Search Login Register
+  MINDMAKERS Forum
|-+  MINDMAKERS.ORG
| |-+  Constructionist AI (Moderator: thorisson)
| | |-+  Modular vs. monolithic architectures
« previous next »
Pages: [1] Print
Author Topic: Modular vs. monolithic architectures  (Read 4946 times)
thorisson
Administrator
Newbie
*****
Posts: 26


View Profile Email
« on: September 13, 2006, 11:55:38 AM »

This is what I have come to the conclusion about regarding modular architectures versus monolithic ones: In the short term monolithic architecture is less work. In the long run, agent-based/modular approaches are less work. The level of difficulty in scaling any architecture increases exponentially with complexiy, so the absolute level of difficulty depends on the difficulty of the subject being modeled. (This is an issue with calibration and if anyone has ideas about how to calibrate this I would love to hear about it.) The increase in scaling difficulty is not as steep for modular architectures as it is for monolithic architectures. In fact, the difficulty of scaling architectures with a potential infinite number of components, where the components don't talk to each other at all, is in a simple linear relationship with the number of components. This is also true for architectures where each component only talks to one other component. As the number of interdependencies between components rises complexity starts to approach monolithic architectures, until at the other end of the spectrum it becomes as tangled and non-transparent.
Logged
thorisson
Administrator
Newbie
*****
Posts: 26


View Profile Email
« Reply #1 on: September 13, 2006, 07:40:45 PM »

Here are some more thoughts:
A good idea in any modular framework is to always try to make the system contain modules with minimal dependency on other modules -- each module should have to know as little as possible about other modules. Dependency between modules has to do with how many modules each module receives information from and how many modules it provides information to. Notice that this goal is separate from the goal of keeping the modules stateless. Statelessness in modules, combined with communication via blackboards, allows centralized recording and managing of global system state (great if you want to recover from a system crash or go back to a previous state) -- i.e. no distributed state (or managed distributed state).
Logged
kennedycolm
Newbie
*
Posts: 6


View Profile WWW Email
« Reply #2 on: November 23, 2006, 08:59:36 PM »

Some of our largest systems today are operating systems and we hear about the same debate there should kernels have a small message passing kernel or should a "delux" kernel cater to every need. IIRC Society of Mind is modular but the problem there is that the messages of the longest reach have to be the simplest.
Logged

-- i luv syms.
Pages: [1] Print 
« previous next »
Jump to:  

Powered by MySQL Powered by PHP Powered by SMF 1.1.4 | SMF © 2006-2007, Simple Machines LLC Valid XHTML 1.0! Valid CSS!