Learning Adaptivity in Heterogeneous Relational Database Systems (LARDS)

With the ever-increasing heterogeneity of hardware, the database community is tasked with adapting to the new reality of diverse systems with a rich set of different architectures, capabilities and properties.
The traditional workflow of hand-tuning implementations to the underlying hardware, for peak performance, is commonly considered untenable for an ever-growing variety of hardware with different performance characteristics. Systems like Micro-Adaptivity in Vectorwise or HAWK have been studied as solutions, but their adoption remains limited.
This project aims to explore solutions for a fully adaptive query execution engine and techniques that allow for simple adoption. To achieve this goal, we plan to tackle four problems.
At first, investigate on how to build micro-optimizations into a hardware-oblivious query pipeline in an efficient and simple-to-maintain way, while still offering a large optimization space. Afterwards, we investigate how to select the best optimizations automatically and in an on-the-fly adapting way, depending on the query and hardware properties.
As a third step, we investigate on the integration of the previous research results into a traditional query execution pipeline and query plan generation.
In the last phase of the project, we will explore techniques that can be used to augment the demonstrator with OLTP capabilities and introduce micro-optimizations into transaction processing.