טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentShmueli Michal
SubjectParallel Management of Events and Situations
DepartmentDepartment of Industrial Engineering and Management
Supervisor Dr. Opher Etzion


Abstract

In recent years, a substantial amount of work has been invested in systems that either react automatically to actual changes (reactive systems), or to predicted changes (proactive systems) in their environment. These systems perform actions or signal alerts in response to the occurrence of events that are signaled when changes in the environment occur (or inferred). Such systems are deployed in a wide spectrum of areas and include command and control systems, active databases, system management tools, customer relationship management systems and e-commerce applications. One of the major issues in these systems is scalability.
In our research we adapt the situation concept of the Amit system, and propose a methodology for achieving an optimized assignment of situations and events to processors according to a pre-selected strategy. The strategy takes into consideration aspects of load balancing methods, and various characteristics of distributed systems, such as communication costs and delays.

Current implementations manage events and situations detection in a sequential and centralized fashion. By proposing this model, we can improve the detection process and provide new capabilities, such as the execution of detection in parallel and a decrease in situation time detection.  This is especially important in real-time systems and in scalable systems, where large volumes of events are handled. The problem of improving the capabilities of active databases has been considered before, usually in the context of rule parallelism, query optimization, coupling modes optimization, but not in the particular context discussed in this work. Our model for solving the problem is a Combinatorial Optimization Problem (COP) model. In this model, N situations have to be assigned to M processors, with constraints. It should be emphasized that this problem is not straightforward since we have to deal with situations that are dependant on other situations (and may be assigned to another processor) events and situations that must preserve a time-dependent order. The predictions of our analytical model have been tested experimentally by simulation and agree with experimental results