The Department of Computer Science at the University of Cyprus cordially invites you to the PhD Defense entitled:

Low-Cost Approximate and Adaptive Monitoring Techniques

Speaker: Mr. Demetris Trihinas
Affiliation: University of Cyprus, Cyprus
Category: PhD Defense
Location: Room 148, Faculty of Pure and Applied Sciences (FST-01), 1 University Avenue, 2109 Nicosia, Cyprus (directions)
Date: Monday, December 18, 2017
Time: 10:00-11:00 EET
Host: Andreas Pitsillides (

With the prevalence of the Internet of Things (IoT) we are starting to see intelligence aggressively deployed at the edge to produce real-time analytic insights for almost all industry sectors. However, to produce such an unprecedented wealth of insights intense processing and constant data dissemination over the network is still required. This results in increased energy consumption for monitoring sources while cloud services consuming IoT data are constantly overwhelmed and struggling to be effective. In this thesis, we tackle real-time data processing and energy-efficiency on the edge of monitoring networks by developing low-cost approximate and adaptive monitoring techniques as the remedy to these challenges. If a degree of inaccuracy can be tolerated, approximate monitoring techniques such as adaptive sampling, filtering and model-based dissemination, can significantly reduce the energy consumption of monitoring sources and the amount of data flooding cloud services by dynamically adapting the metric stream collection and dissemination rate. To achieve this, we introduce low-cost approximate and probabilistic algorithmic learning models that capture runtime knowledge from the metric stream evolution and variability, adjusting the collection and dissemination rate based on the confidence of each algorithmic model to correctly estimate what will happen next in the metric stream. Specific consideration is taken to fine-tune each algorithmic model at runtime by introducing adaptive parameter weighting, trend detection and seasonality behavior enrichment so that our algorithms immediately identify abrupt transient changes in the metric stream evolution and overcome any lagging effects in the estimation process. Next, we proceed by introducing AdaM, a lightweight framework embeddable in the software core of monitoring sources (e.g., IoT devices) that provides model-based adaptive monitoring by incorporating the low-cost approximate monitoring techniques developed in the scope of the thesis. We conclude with a thorough experimentation study of AdaM using real-world data from cloud applications, wearables and intelligent transportation services. Results show that AdaM is able to achieve a balance between efficiency and accuracy, and particularly, it can reduce energy consumption by at least 83%, data volume by 71%, while maintaining accuracy always above 90% in comparison to other state-of-the-art adaptive frameworks.

Short Bio:
Demetris Trihinas is a PhD Candidate at the Computer Science Department of the University of Cyprus. He additionally holds a Computer Science MSc from the University of Cyprus and a Dipl-Ing in Electrical and Computer Engineering from the National Technical University of Athens (NTUA). His research interests include Distributed and Internet Computing with particular focus on Cloud Monitoring and Elasticity, and the Internet of Things. He is the developer of the JCatascopia cloud monitoring system, the AdaM framework for IoT devices and is also a member of the Cloud Application Management Framework (CAMF) which is now an official Eclipse IDE project. Additionally, Demetris is a full-time researcher at the Laboratory for Internet Computing (LInC) of the University of Cyprus and is currently contributing to the Unicorn project which is co-funded by the EU commission H2020 framework. His work is published in IEEE/ACM journals and conferences such as INFOCOM, CCGrid, BigData, TCC and ICSOC, ICWE and EuroPar.

  Other Presentations Web:
  Colloquia Web: