Koji Kida

A scalable complex event processing system and evaluations of its performance

This paper describes a scalable context delivery platform (SCTXPF) and our evaluations of it. The SCTXPF receives a large number of events from various event sources and a large number of complex event processing (CEP) rules from various services/applications. The SCTXPF achieves load distribution of CEP operations by parallelizing event processors (EPs) and allocating CEP rules to each EP. The SCTXPF should allocate CEP rules efficiently to be able to operate with a high level of performance and be scalable. The rule allocation algorithm allocates the CEP rules to EPs so that the state of event processing is efficiently managed. The SCTXPF achieves high throughput and scalable CEP with the EPs operating independently of one another. We evaluated the efficiency of the rule allocation algorithm in terms of scalability and performance. The results revealed that the proposed system is scalable and the performance reached 2,700,000 events/sec and the proposed algorithm improves its scalability.

A complex event processing for large-scale M2M services and its performance evaluations

This paper describes a scalable complex event processing (CEP) system and rule allocation algorithm for large-scale M2M services and presents the results of its performance evaluations. Platforms providing M2M services deal with various data from massive event sources in the real world. To realize these services, a scalable system is required. Our previous system determines efficient allocation of CEP rules based on a single key attribute type. However, it cannot maintain scalability in M2M situation because CEP rules are defined by various attribute types and there is no single key attribute, i.e. common attribute type among CEP rules. The challenging issue in the present work is how to maintain scalability where no common attribute type usable for rule allocation. The key idea in this paper is to allocate CEP rules based on different attribute types for each CEP rule instead of single key attribute type. We propose a rule allocation algorithm that determines the efficient allocation of CEP rules when the system needs processing a large number of events, a large number of CEP rules, and various types of events. Evaluation results show that the system with the proposed algorithm is scalable with many CEP Servers. We thus conclude that the proposed algorithm contributes to the realization of large-scale M2M services.