The Commercial Application of a TeaTime Inspired Distributed Computing Framework

KSX is an expert system development environment for advanced process control applications. It uses a English-like rule language with integrated fuzzy logic, neural network models and genetic algorithm optimizers. In order to adapt to the move towards ubiquitous multi-core and cloud environments, a new design for distributed computing was required. Inspired by Croquet’s TeaTime architecture we created a distributed computing framework for KSX using Cincom’s VisualWorks and Opentalk. This framework uses Opentalk’s multicast for presence, and Opentalk’s TCP based Smalltalk to Smalltalk messaging.  Some of the features of the system as a result of this framework are:  Remote client connections Multiple client collaboration Parallelization of expensive tasks like model training, or costly genetic algorithm fitness functions Increased fault tolerance by “sharding” application logic Creation of hierachical control strategies using “portals” or connections between worlds  In this talk I will present the mechanics of the framework, demonstrate the KSX application using a distributed genetic algorithm and comment on some of the design challenges and decisions we faced when implementing this system in our commercial application. Some of the features of the system as a result of this framework are:  Remote client connections Multiple client collaboration Parallelization of expensive tasks like model training, or costly genetic algorithm fitness functions Increased fault tolerance by “sharding” application logic Creation of hierachical control strategies using “portals” or connections between worlds  In this talk I will present the mechanics of the framwork, demonstrate the KSX application using a distributed genetic algorithm and comment on some of the design challenges and decisions we faced when implementing this system in our commercial application.

Scribd