Accord: Accelerating Big Data Applications on Heterogeneous Processors

Many applications require fast access, processing, and analytics of their big data, for example, millions of image files, tables containing hundreds of millions of rows, and billion-edge graphs. With the rapid advances of hardware technologies, these applications run on powerful servers with heterogeneous processors, computer clusters, and cloud computing platforms. However, the data handling and processing is often not in accordance with the underlying hardware resources. Therefore, we proposed to develop software techniques that accelerate these applications by utilizing their hardware resources effectively. In particular, we work on data-parallel primitives as well as parallel algorithms on graph analytics and relational queries. The processors we consider include multicore CPUs, Intel Knights Landing (KNL) processors, and GPUs.


Source Code Repository

Rapids at HKUST on GitHub


We thank our industrial sponsors Alibaba and SenseTime for their generous support.