About
I am a PhD student in the Database Systems group at TU Munich, advised by Prof. Jana Giceva. My research sits between database systems and cloud infrastructure. Cloud warehouses promise elasticity, but every design choice (how to provision, how to handle metadata, when and how to scale) comes with a cost that is hard to reason about. I build models, simulators and benchmarks that make those trade-offs explicit, and then use them to argue for and implement better system designs.
Lately I have also been looking below the database layer: how GPUs behave when their memory is oversubscribed, and what memory-centric hardware means for how we should architect data systems.
News
- Oct 2026Ember, our preprint on making Unified Virtual Memory practical for oversubscribed GPUs, is on arXiv.
- Oct 2026Presented CloudGlide at the Bavarian Database Day in Munich.
- Oct 2026Co-teaching the seminar Next-Generation Data Systems this winter.
- Sep 2026Presented CloudGlide at VLDB 2026 in Boston.
- Jun 2026Two papers at SIGMOD 2026: the CloudGlide demo and our cloud benchmarking paper at DBTest.
Research
-
Cloud database architectures
How do we build database systems for the cloud that get the most out of object storage, run analytics cost-efficiently, and scale transactions across regions?
CloudGlide PVLDB '25 CloudGlide demo SIGMOD '26 World Wide Wait ongoing
-
Metadata management & caching
Metadata can bridge the gap between cold and hot data access, but managing it at scale raises its own questions: which metadata to maintain, and where to keep it.
Ongoing work.
-
Modern hardware & efficient computing
Enabling larger and more resource-efficient data processing on GPUs through unified memory, and our vision for how memory-centric hardware should be designed.
Ember arXiv '26 · ongoing Memory-centric database architectures Datenbank-Spektrum
-
Serverless query processing
Serverless comes closest to the pay-as-you-go promise of the cloud and scales to very large data processing, but every unit of resource costs more. Query processing has to trade cost against latency explicitly.
Multi-objective optimization in Spark SQL IDEAS '22 Cost-performance optimization in serverless query processing ongoing
-
Benchmarking
TPC-H says little about real cloud workloads, and vendor telemetry is too noisy to rebuild workloads from. We argue for a clearer definition of transferable workload features, and for benchmarks built from concrete use cases.
Cloud analytics benchmarking DBTest '26 · ongoing
Publications
2026
-
Making Unified Virtual Memory Practical for Oversubscribed GPUs
-
Rethinking Database Architectures for Memory-Centric Computing
-
Cloud Analytics Benchmarking: Where We Are, What Is Missing, and Where We Should Go
-
Demonstrating CloudGlide: Interactive Exploration of Cloud Analytics Architectures
2025
-
CloudGlide: Deconstructing the Landscape of Cloud-Based Analytics
2023
-
Towards Next-Generation Resource Optimization in the Cloud
2022
-
From Cloud to Serverless: Multi-Objective Optimization in the New Cloud Epoch
-
Multi-Objective Query Optimization in Spark SQL
† equal contribution
Background
- 2022 –PhD in Computer Science, TU MunichChair for Database Systems, advised by Prof. Jana Giceva
- 2024Software Developer, Firebolt AnalyticsScheduling for distributed query execution in a cloud data warehouse (Jul – Oct)
- 2022Software Engineer, Qubit Systems, AthensBackend development and analytics (Feb – Jul)
- 2016 – 21M.Eng. in Electrical and Computer Engineering, National Technical University of AthensThesis: Multi-objective Query Optimization for Massively Parallel Processing in Cloud Computing
Teaching
Teaching assistant at TUM since 2022, mostly on the cloud and hardware side of the curriculum.
- WS 26/27Seminar: Next-Generation Data Systemsco-organizer
- SS 2026Praktikum: Cloud-Based Data Processinglead organizer
- WS 25/26Advanced Algorithms in Competitive Programming
- SS 23 – 25Data Processing on Modern HardwareSS 2023 · SS 2024 · SS 2025
- WS 22 – 25Cloud-Based Data ProcessingWS 22/23 · WS 23/24 · WS 24/25
- SS 2023Seminar: Modern Database Systems
Lectures
- WS 25/26Distributed Query ProcessingFoundations of Data Engineering
- WS 24/25Cloud-Based OLAP SystemsCloud-Based Data Processing
- WS 23/24Data CentersCloud-Based Data Processing
Service
Students
Ongoing
- MTWorld Wide Wait: Exploring the Design Space of Geo-Distributed OLTP Database SystemsBorbala Fazakas cloud
- MTA Hardware-Based Prefetcher for Stabilizing GPU Unified Memory Performance under OversubscriptionSenad-Leandro Lemes Galera hardware
- BTShared Caching for Cloud OLAP Systems: When Is It Worth It?Artem Lomov caching
Completed 9 theses
- GR 2026Cost-Performance Optimization in Serverless Query EnginesBorbala Fazakas serverless
- GR 2026Stabilizing Discrete GPU Unified Memory Performance under OversubscriptionSenad-Leandro Lemes Galera hardware
- BT 2026Designing an Interactive Visualization Tool for Cloud-Based Analytics SimulationsLukas Weingand cloud
- BT 2026Kernel-Based Hypervisor for Lightweight VMsThomas Wachter hardware
- MT 2025Designing a Unified In-Memory Caching Layer for Cloud OLAP WorkloadsGregor Caf caching
- MT 2024Workload-Driven Optimization of Autoscaling Systems Using Heterogeneous NodesMatthias Fesl cloud
- MT 2024Syslog Collection and Analysis of Network Devices in Microsoft AzureCamelia Zalum cloud
- GR 2023Quantifying the Noisy Neighbor Problem in the CloudChristodoulos Lykourinou cloud
- BT 2023A GUIde for Tuning and Troubleshooting Parallel Query ProcessingVincent Limbach cloud
BT: Bachelor's thesis · MT: Master's thesis · GR: Guided Research
Beyond research
I grew up in Athens. Outside work I like to play guitar, dance lindy hop, and go to the cinema.
I watch football and have been waiting my whole life for Panionios to become good.
I also like to read literature, learn about history and geography, and then test that knowledge, unsuccessfully, at pub quizzes. Occasionally I edit videos.