The expected performance benefits of offloading some important basic database operations to computational storage.
This paper discusses the expected performance benefits of offloading some important basic database operations — namely Scan, Filter and Project — to computational storage. We evaluate the performance estimate model using TPC-DS workload and two database engines running on Hadoop clusters: SPARK- SQL and Presto.
This paper is organized as follows: after covering previous computational storage database offloading work, we explain the OLAP workload selection, and the configuration of our two clusters. In Section IV we dive into TPC-DS characteristics and examine the overall performance from running on the two Hadoop clusters, which have been the focus of our experimentation. In Section V, we explain our modeling methodologies, and in Section VI we describe and analyze results from that modeling. Specifically, we show how a substantial speed-up from computational storage optimization can depend on multiple factors. Finally, we briefly discuss other SQL building blocks amenable to computational storage pushdowns, and conclude.
Click here to read more.
 |
 |
 |
 |
 |
 |
 |  |
Leave a Reply