چكيده به لاتين
Abstract
The growing tendency to event-based architectures on the one hand and the prevalence of computing systems on the other hand, has caused the increasing growth of event-producing resources and consequently the rise in event production rate. In many domains of application, it is necessary for the computing systems to acquire the required understanding of their environment by processing events, recognizing the predefined patterns and taking them to higher levels of abstraction, and to show a certain reaction after taking the right decision. This type of processing is called complex event processing. The escalation and spread of application of such systems, in addition to causing high event processing rate, has transformed the need to recognize multiple patterns, which are mostly very complex, to a major challenge. Centralized approaches, cannot give an appropriate performance, due to the formation of a bottleneck and eventually the lack of ability to provide the required resources for processing events. These approaches encounter serious challenges, such as the inability to recognize all the complex events and delay in recognizing them. Therefore, using multiple computing resources in order to deal with high volumes of events and variety of patterns, is inevitable. Our proposed strategy in this research is to provide a mechanism for distributing patterns among computing nodes and using the computing power of several computing resources for processing events and recognizing these patterns. In this mechanism, by using the bin-packing algorithm and based on the estimation of the processing unit and the amount of memory each rule needs, and also the amount of resources available in the processing units, the rules are distributed among these nodes in a way that computational load and consumed memory is as balanced as possible. In fact, by applying this mechanism, we try to prevent inappropriate behaviors like the inability to serve in a node due to lack of resources in it. The results of our research show that using this method of pattern distribution causes the increase in throughput and reduction in the number of unrecognized complex events.
Keywords: Complex event, Complex event processing, Events stream, Rule distribution, Parallel processing
