Hyper and In-Memory Databases
AWS has made it easy to get the latest and greatest hardware. You can easily get machines O(TB) of DRAM for a few dollars per hour. r5.16x large instance for example has the following specs:
r5.16xlarge 64vCPU 256ECU 512 GiB EBS Only $4.032 per Hour
TPC-H benchmark is a standard dataset is a popular OLAP benchmark. Scale factor 1000 of TPC-H benchmark generates 1TB of data which after removing ununsed columns and dictionary compression come to nearly 280GB of data. This dataset has 6 billion rows ! This dataset can easily be operated on using an in-memory database — one of the best in-memory databases is Hyper. Hyper can operate at memory-bandwidth speed (~150GBps on a Xeon today).
Hyper could processs queries on this dataset sub 2s for aggregation queries and sub 10s for more complex SQL queries. I find it surprising that today when most companies (all except big tech) likely have lesser than 6 billion rows and use things like Redshift / Snowflake to run their workloads. They could make their workloads like BI dashboards, ad-hoc analysis, etc much faster by simply using an in-memory database.
To try Hyper, download Tableau Server distribution and install it. You don’t need a license to run Hyper. Once installed:
# cd to Tableau install dir/packages/hyper.<ver>/
./hyperd --log-dir . -d db --init open --skip-license --no-password configure.
./hyperd --log-dir . -d db --init open --skip-license --no-password run
# connect using psql
psql -h localhost -U root -p <port>
The instructions work on Tableau Server 20.3.