A scalable grid based application platform for high volumes of transactional event driven processes

Events are a technical representation of facts. A fact is anything that can increase the entropy of the universe. A telephone call. An object changing its position. A network packet crossing a router. Facts can manifest in chaotic clouds and are not necessarily correlated. Of particular interest are the facts that can be sensed and that are useful within a business environment. The technology described in this paper is based on the following principles:

a) each fact of interest is always related to a sensor providing a first technical representation of the fact, i.e. an event

b) fact emitters are always uniquely identifiable

c) a fact can have an impact on an object with a scope wider than the scope of the reaction to the fact

d) the reaction applicable to a fact might or might not depend on the status of objects impacted by facts occurring along a timeline

e) if a fact leads to a reaction modifying the status of an object with a wider scope than that of the reaction to the fact, then it is possible to establish a relationship between the fact emitter and such an object.

The application of these principles results in a new methodology for the design of event-driven business processes, based upon a ternary associative logic between event emitters (or agents), reactions to events (i.e. processes) and their impact on global objects. This paper discusses the technology and architectural requirements for the execution of such event-driven processes.

The result is a distributed, multi-process and multithreaded event driven architecture based on an application platform providing, as built-in services, scalability at both the execution and the persistence layers, together with contention resolution and transaction co-ordination. This paper presents scalability and performance benchmarks for such event-driven application platform (EDAP). Using as benchmarks process a simple rating engine, the authors have observed a sustainable performance of up to 11,600 statements (read and write operations against DBs) per second on a small grid of inexpensive computers.