| Title | Authors | Synthesis | Publisher | Keywords |
| Monkey: Optimal Navigable Key-Value Store | Niv Dayan, Manos Athanassoulis, Stratos Idreos | This paper presents Monkey. The paper gives the formal mode for the look-up, update and range query cost. The model is built on the merge policy, the write buffer size, and the false posstive rate for bloom filters. An algorithm is invented to find the optimal parameter setup for LSM-DB and given workload. | SIGMOD 2017 | |
| SageDB: A Learned Database System | Tim Kraska, etc | SageDB is a database system that uses leanred models, for example RMI based models, to optimize database system components. This paper expand the applications of learned models on database indices. It shows learned models can be used to estimate optimizer models, improve joins and sorting operations, task scheduling and mutliple dimensional indices. | CIDR 2019 | |
| The Periodic Table of Data Structures | Stratos Idreos, etc | This paper presents the design space of Data Structures. The authors call it the Periodic Table of Data Structures. This paper distills the first principles of data structure design from design choices of data layouts. And this paper shows that by interpolating this design principles with current designs we can get new data structures. And with a technique called Learned Cost Model it is possible to infer the performance by only implementing a small part of the target data structures. | IEEE Data Eng 2018 | |
| H20 A Hnads-free Adaptive Store | Ioannis Alagiannis, Stratos Idreos, Anastasia Ailamaki. | This paper presents an adaptive database system which can analyze the workload and find the better columon configuration. This system tries to solve the fixed storage layout problem. Starting from a pure column store H20 tries to optimize the storage layout by merging existing columns. Through experiment this system can adjust to changing workloads. | SIGMOD 2014 | |
| Database Cracking | Stratos Idreos, etc | This paper presents a technique called database cracking which is similar to index building driven by queries. Unlike ordinary systems indices are built explicitly by users or by background optimizers. Database cracking postpones the indices building when it is found that it helps the workload. And it does not fully order the data, but only to the extent that required by the query. | CIDR 2007 |