GPUs have been successfully used to accelerate workloads in a number of fields of research. In data management, efforts mainly have been directed at bringing query processing and OLAP to GPUs.

In this paper, we argue that not OLAP but OLTP workloads should be the main target for GPUs. As a main contribution we thus present GalOP — a GPU-accelerated OLTP main memory DBMS. At the core GalOP is based on a novel deterministic concurrency scheme for GPUs which orders conflicting transactions before the execution on the GPU and thus makes OLTP more robust for high and low conflict scenarios. In our initial evaluation, we show that GalOP can provide robust performance for low- and high-conflicting workloads and outperforms a pure CPU-based scheme by up to 10x.

Authors:

• Nils Boeschen (TU Darmstadt)
• Carsten Binnig (TU Darmstadt)