Unit:
Department of Informatics and TelecommunicationsΠληροφορική
Author:
GRIGORIOU EVANGELOS
Supervisors info:
Δελής Αλέξης
Βαθμίδα: Καθηγητής
Τμήμα Πληροφορικής και Τηλεπικοινωνιών
Εθνικόν και Καποδιστριακόν Πανεπιστήμιον Αθηνών
Original Title:
An Efficient Decentralized Streaming Model
Translated title:
An Efficient Decentralized Streaming Model
Summary:
Recently, increasingly large amounts of data are generated from a variety of sources. Streaming frameworks for Big Data applications help to store, analyze and extract useful informationfromsuchcontinuouslygenerateddata.Thereareseveralexistingstreaming frameworks, like Apache Storm, Apache Spark and Apache Flume. Inthisthesis,wepresentadecentralizedstreamprocessingmodel.ItusesaDHTprotocol toachieveamanymasters-manyworkersarchitectureandassigneachjobitsownmaster. Evengroupsarecreatedforeachjobbyutilizingthesystem’sroutingproperties,resulting in a hierarchical tree formation, consisted of agents that are participating in the network. The root of this tree acts as the master of the group and is responsible for synchronizing the group’s members. Each agent consumes live data logs, which are parsed into mini batches and stored in a memoryefficientdatastructure.Theagentsaggregatetheirlocaldataandtheresultsare rolled up the the aggregation tree.
Main subject category:
Science
Keywords:
big data, cross job coordination, compressed buffer tree, distributed systems, many masters architecture
EvangelosGrigoriou-Thesis.pdf (270 KB)
Open in new window
EvangelosGrigoriou-Src.zip
90 KB
File access is restricted.