Research

Overview

For almost eighteen years our research topic has been "Fast access to large data". You may call this "big data", you may call it "massive datasets", you may call it "very large data". Whatever. In any case, we are interested in providing query processing techniques allowing you to manage very large datasets.

Current Topics

• Deep Query Optimization
  Query optimization is at the heart of every database system and enables fast and efficient query processing. State-of-the-art query optimisation includes the translation of a logical plan to a physical plan on the granularity of complete operators. We are currently working on deepening this process by breaking up the abstraction of a (physical) operator to consider more fine-granular subcomponents. With that, we aim to apply many more optimizations which would otherwise be missed.
  
  
• mutable
  Our goal with mutable is to develop a modular database system that allows for fast prototyping of research ideas. The design should enable abstraction such that high-level ideas and algorithms can easily be implemented into the system. At the same time, we leverage modern concepts of software design to achieve high performance. We do not want to trade off between abstraction and performance, instead we work towards the best of both worlds.
  
  
• Blockchain
  
  
• Applied Machine and Deep Learning
  What is the impact on ML and DL to databases and big data technology? We are both interested in applying ML/DL to real-world problems as well as leveraging DB-technology to improve ML and DL methods. JD gave a keynote on this topic at VLDB 2017 and will give another keynote at the DEEM-workshop (Data Management for End-To-End Machine Learning) at SIGMOD 2018. We have several ongoing projects in that space such as weather prediction using deep learning and DeepTuneDB/NoDBA and also a start-up: d:AI:mond.ai.
  
  
• Indexing
  indexing is key to providing efficient query processing in almost any information system. We are interested in problems that may not be solved with existing techniques or require a fresh look at available techniques. We are especially interested in adaptive indexing techniques (adapting to query/workload patterns) and indexing support under high update rates (millions of updates per second). We won a VLDB 2014 best paper award for this work.
  
  

Past Topics

• Hyperfast Snapshotting
  How can you manipulate virtual memory to speed-up snapshotting vor concurrency control and other applications. One effort in that space is RUMA. An extension of this work is a slight extension to a linux-kernel making this even faster.
  
  
• new system architectures for data management
  The idea of a Database Systems was developed more than 30 years ago. Meanwhile many new data managing problems have been identified. Several new data models were invented. Moreover hardware has changed dramatically. We doubt that the abstraction of a Database Management System is always the best abstraction. Therefore we are interested in coming up with better system abstractions that cover a wider range of information management problems. One effort in that direction is our project OctopusDB.
  
  
• main memory and flash databases
  Many databases today already fit into main memory. Systems are going away from using disks to using flash and DRAM memory. In addition, new kind of storage media is coming up, e.g., persistable RAM. This new hardware will affect how information systems are built.
  
  
• data warehousing and OLAP
  Online Analytical Processing (OLAP) is at the heart of any medium to large business. OLAP systems provide instantaneous answers to complex queries. Building these systems is far more challenging compared to 'standard' OLTP systems.
  
  
• MapReduce/Hadoop
  MapReduce/Hadoop is a popular paradigm for large-scale data analysis. However, typical MapReduce tasks suffer from performance issues. We look at ways to improve runtimes. One effort in that direction is our project Hadoop++ and its successor HAIL (Hadoop Aggressive Indexing Library).
  
  
• dataspaces
  Dataspaces are a new abstraction for information management. A Dataspace system is a new kind of information integration system that incorporates features of search engines as well as EII/OLAP/DWH. We have built one of the first dataspace management systems: iMeMex. It uses declarative trails to enrich a dataspace. video
  
  
• moving objects and vehicle tracking including cars and aircraft
  Indexing moving objects refers to scenarios where the data is changing quickly, e.g., cars, trains, aircrafts change their positions continuously. How do you provide scalable tolling, tracking systems for these kind of applications? The MOVIES project investigates these research questions. In that context we also developed a trace generator to simulate large datasets of moving objects called moto. This simulator overcomes scaling issues prevalent in existing generators.
  
  
• personal information management
  Personal information refers to all files on all your devices (including your laptops, phones, iPods, audio systems, etc.). Currently people use a zoo of techniques to manage and query this information. We look at ways of coming up with a unified system for personal information management allowing you to handle your data with a single system.

If you are from industry and are interested in working with us in of the above areas, please feel free to contact us.