Ημερολόγιο Εκδηλώσεων

Περιγραφή:
ΠΟΛΥΤΕΧΝΕΙΟ ΚΡΗΤΗΣ Σχολή Ηλεκτρολόγων Μηχανικών και Μηχανικών Υπολογιστών Πρόγραμμα Μεταπτυχιακών Σπουδών ΠΑΡΟΥΣΙΑΣΗ ΜΕΤΑΠΤΥΧΙΑΚΗΣ ΕΡΓΑΣΙΑΣ MΑΝΙΚΑΚΗ ΒΑΣΙΛΙΚΗΣ με θέμα Αρχιτεκτονική και υλοποίηση ενός συστήματος επεξεργασίας πολύπλοκων γεγονότων Architecture and Implementation of a Distributed Complex Event Processing System Τρίτη 4 Απριλίου 2017, 1 μ.μ. Αίθουσα Συνεδριάσεων Softnet, Κτίριο Επιστημών, Πολυτεχνειούπολη Εξεταστική Επιτροπή Αναπληρωτής Καθηγητής Αντώνιος Δεληγιαννάκης (επιβλέπων) Καθηγητής Μίνως Γαροφαλάκης Επίκουρος Καθηγητής Βασίλειος Σαμολαδάς Abstract Distributed event detection is the process of identifying specific occurrences of interest in incoming data available at a number of distributed nodes. The traditional approach for detecting events implies central collection and processing of data, which is impractical for a number of reasons. Firstly, since the number of nodes might be large, collecting information centrally is not always possible or efficient. This happens because the amount of information to be transmitted may be huge and the available bandwidth insufficient to accommodate the transmission. Secondly, central processing of distributed data is not balancing the cost for answering more complex queries and for big data applications the processing speed may introduce additional latency in complex event detection. Additionally, processing all data in a distributed network in a single node generates a single point of failure. In-situ processing for complex event detection systems is an architectural scheme that can alleviate the aforementioned limitations. It provides a mechanism for balancing the work load of both event processing and network traffic by distributing coordinating duties to multiple nodes. The additional division of monitoring complex queries in multiple steps, based on event frequencies, enables each node to relay just the absolutely required events to the coordinating nodes for evaluation. Additionally, the geometric method allows a network to monitor in a distributed way if the value of a complex function, even nonlinear, calculated using incoming data is over or under a specific threshold value. Thus, composite events can be distributedly detected if they are expressed as a threshold monitoring function. The geometric method imposes a set of local constraints on each node and manages to reduce the need for communication between the nodes as long as the constraints are satisfied. In this work, a unified architectural integration of in-situ complex event processing and the geometric method is implemented, using the real-time distributed computation framework named Storm, for distributed event detection. A topology is implemented to handle both the monitoring of complex functions as well as complex event queries using Storm components. All necessary mechanisms for intra- and inter-node communication are also addressed to facilitate the optimization objectives. Finally, the system is designed to recover after a node transient failure and special care is taken to allow real-time system adaptivity in case event frequencies drift significantly over time.