Past Projects
Distributed Tiered Storage for Cluster Computing
Improvements in memory, storage devices, and network technologies are continually leveraged by distributed systems to meet the increasing data storage and I/O demands of modern large-scale data analytics. We present OctopusFS, a novel distributed file system that is aware of storage media (e.g., memory, SSDs, HDDs, NAS) with different capacities and performance characteristics. The system offers a variety of pluggable policies to automate data management across both storage tiers and cluster nodes. A new data placement policy employs multi-objective optimization techniques for making intelligent data management decisions based on the requirements of fault tolerance, data and load balancing, and throughput maximization. Moreover, machine learning is employed to track and predict file access patterns, which data movement policies then use to decide when and which data to move up or down the storage tiers to improve system performance. This approach uses incremental learning with XGBoost to dynamically refine the models as new files are accessed, thereby improving their predictive performance. At the same time, the storage media are explicitly exposed to users and applications, allowing them to choose the distribution, placement, and movement of replicas within the cluster based on their performance and fault-tolerance requirements.
Real-time Aggression Detection on Social Media
The rise of online aggression on social media is becoming a major concern. Several machine learning and deep learning approaches have been proposed recently for detecting various types of aggressive behavior. However, social media is fast-paced, generating an increasing amount of content, while aggressive behavior evolves over time. We introduce the first practical, real-time framework (RADONS) for detecting aggression on Twitter by embracing the streaming machine-learning paradigm. The framework is designed to be adaptable (its ML classifiers are trained incrementally as they receive new annotated examples), scalable (it can process the entire Twitter Firehose with three machines), and generalizable (it can detect other abusive behaviors such as sarcasm, racism, and sexism in real time). This project was carried out in collaboration with Dr. Nicolas Kourtellis from Telefonica Research, Spain, and Dr. Despoina Chatzakou from the Center for Research and Technology Hellas, Greece.
Scaling Transactional Databases with Strong Guarantees
Database replication is a common mechanism for scaling performance and improving the availability of transactional databases, but past approaches have suffered from various issues, including limited scalability, performance-versus-consistency trade-offs, and requirements for database or application modifications. Hihooi is a replication-based middleware solution that provides workload scalability, strong consistency guarantees, and elasticity for existing transactional databases at low cost. A novel replication algorithm enables Hihooi to asynchronously propagate database modifications to all replicas at high speed while ensuring consistency across replicas. At the same time, a fine-grained routing algorithm is used to load-balance incoming transactions across available replicas in a consistent way. This project was carried out in collaboration with Dr. Michael Sirivianos at the Cyprus University of Technology.
MERMAID: Mediterranean Marine Eutrophication Diagnostics using Artificial Intelligence
Cyprus's coastal waters are highly important to both tourism and the rapidly growing aquaculture industry, making the monitoring of eutrophication essential for protecting marine ecosystems and ensuring compliance with environmental regulations. The MERMAID project developed and adapted deep learning models for marine areas around Cyprus using in situ measurements of key water quality parameters, including temperature, salinity, nutrient concentrations, dissolved oxygen, and electrical conductivity. Self-organizing maps were employed to investigate relationships between environmental variables and chlorophyll a concentrations, thereby confirming the oligotrophic character and good ecological status of Cypriot coastal waters. In addition, feed-forward neural networks were used to predict chlorophyll a, dissolved oxygen, and dissolved inorganic nitrogen levels, while sensitivity analysis identified the most influential environmental factors and captured seasonal phenomena such as winter upwelling. The resulting models provide valuable decision-support tools for eutrophication management, hypoxia prevention, and the early detection of harmful algal blooms.
SCURMEM: Sea Current Modeling in the Eastern Mediterranean Sea
The SCURMEM project focuses on developing an operational oceanographic monitoring and forecasting system for the Eastern Mediterranean, with particular emphasis on the waters surrounding Cyprus. Accurate estimates of sea surface currents are essential for applications such as oil spill tracking, search-and-rescue missions, pollution monitoring, and environmental risk assessment. The research began with a comparative evaluation of several established hydrodynamic models and data products, including CYCOFOS, MITgcm, FVCOM, and the Copernicus Marine Environment Monitoring Service (CMEMS), to assess their suitability for regional deployment. The project then integrated numerical simulations with in-situ observations from ships via Automatic Identification System (AIS) data and from autonomous multi-sensor floating platforms. By combining these data sources with dead-reckoning techniques, the system improved the accuracy of sea-current predictions, particularly in coastal regions. The resulting platform provides a valuable decision-support tool for environmental protection, maritime safety, and marine resource management in Cyprus.
Maritime Cognitive Decision Support System
The primary objective of the MARI-Sense project was to integrate and adapt existing expertise and develop novel knowledge and skills to build the MARI-Sense Cognitive Decision Support System for Maritime Activities Planning, Emergency Response and Planning, and Maritime Spatial Planning. The secondary general objective was the development and implementation of strategies for smart, sustainable, and inclusive growth with a beneficial impact on society, technology, and the economy, powered by the diverse capabilities of members of the quadruple helix and the general public. The project was co-funded by the European Regional Development Fund and the Republic of Cyprus through the Research and Innovation Foundation (RIF) with a total budget of 1M Euros.
Sea Traffic Management in the Eastern Mediterranean
The general objective of the STEAM (Sea Traffic Management in the Eastern Mediterranean) project was to efficiently manage sea traffic in the Eastern Mediterranean while ensuring safety and environmental sustainability. More specifically, to develop the Port of Limassol to become (i) a world-class transshipment and information hub adopting modern digital technologies brought to the maritime sector, and (ii) a driver for short sea shipping in the Eastern Mediterranean through enhanced services based on standardized ship and port connectivity. The project was co-funded by the European Regional Development Fund and the Republic of Cyprus through the Research and Innovation Foundation (RIF) with a total budget of 1M Euros.
Middleware for Combining Heterogeneous IoT Devices
The Internet of Things (IoT) extends connectivity beyond traditional computing devices to a wide range of smart objects equipped with various sensors and actuators. Due to device heterogeneity, developing applications that require collecting and sharing data across multiple IoT devices is complex, as developers must be familiar with a diverse set of supported services and APIs. Cuttlefish is a flexible, lightweight middleware that provides a unified API to support the development of applications that integrate multiple heterogeneous IoT devices. It abstracts away much of the complexity involved with orchestrating different devices at runtime. At the same time, it avoids the caveats of existing approaches through a simple and efficient design, while offering a rich set of capabilities to developers. This project was carried out in collaboration with Dr. Andreas Pamboris at UCLan Cyprus.
Towards a Unified Platform for Multi-Wearable Apps
Wearable technology has recently become an ubiquitous part of everyday life. Smartwatches, activity trackers, and sensor-embedded clothing are used to monitor personal fitness data, detect health disorders, and provide real-time feedback. However, the current landscape of wearable devices suffers from two main issues: (i) each device currently offers only a portion of all the combined capabilities of all the devices, and (ii) most devices do not share data with each other and are tied to certain ecosystems. Hence, there is a strong need for a unified framework that will transform the current collection of standalone devices into a fully networked system. This project was carried out in collaboration with Dr. Andreas Pamboris and Dr. Panayiotis Andreou at UCLan Cyprus.
Sea Traffic Management Validation Project
The primary goal of this research program was the innovative optimization of processes and services within and between ports based on enhanced collaboration and regulated information sharing among port actors. Sea Traffic Management will create significant added value for the maritime transport chain, in particular for ship and cargo owners. This was achieved through real-time, efficient information exchange among various parties, such as ships, service providers, and shipping companies. The project was funded by the European Commission, Innovation and Networks Executive Agency (INEA), Connecting Europe Facility (CEF), with a total budget of 42.9M euros.
Scalable Near Real-Time Failure Localization of Data Center Networks
Despite the built-in redundancy in data center networks, performance issues and device or link failures in the network can lead to user-perceived service interruptions. Therefore, determining and localizing user-impacting availability and performance issues in the network in near real time is crucial. Our key idea is to apply statistical data mining techniques to large-scale active monitoring data to produce a ranked list of suspect causes, which we refine using passive monitoring signals.
Starfish: A Self-tuning System for Big Data Analytics
The Hadoop MapReduce platform is a popular choice for big data analytics. Unfortunately, Hadoop's out-of-the-box performance leaves much to be desired, leading to suboptimal use of resources, time, and money. Starfish is a self-tuning system for big data analytics that builds on Hadoop, adapting to system workloads and user needs to deliver good performance automatically, without requiring users to understand and manipulate the many tuning knobs in the Hadoop platform.
Query Optimization Techniques for Partitioned Tables
Table partitioning has evolved into a powerful mechanism, but is currently not utilized effectively during query optimization. We have developed new techniques to generate efficient plans for SQL queries involving multiway joins over partitioned tables. The techniques are designed for easy incorporation into bottom-up query optimizers and have been prototyped in PostgreSQL.
Automating the Process of SQL Tuning
zTuned is a new system that automates SQL tuning using an experiment–driven approach. The nontrivial challenge is to plan the best set of experiments to conduct so that a satisfactory (new) plan can be found quickly. A novel feature of zTuned is a SQL-tuning-aware query optimizer, called Xplus, that proactively executes plans, collects monitoring data from runs, and iterates. Xplus has been prototyped using PostgreSQL.
