Follow the steps described to build the code across a cluster
Follow the steps described to run the cluster.
These steps are for the deployment of a cluster all on a single node (two data sites, and an advisor). Deploying across a cluster is similar, just with changes to where the scripts execute, and changing hostname/mport pairs in config files.
By changing the benchmark configuration file and benchmark name to oltpbench, different benchmark workloads can be executed.
Additionally, by changing the global configuration file to different constant values from the default, you can change the parameters of the system. For instance, changing SPAR hyperparameters
The experiments with cracking can be run by running the cracking driver, after building the source code.
Changing the gflags passed to the driver changes the experimental setup. For example, to run an experiment with only the SPAR predictor, run: ./cracker --scrack_do_prediction --scrack_eqp_type=SPAR
The scripts assume a certain configuration of repositories and directories. Moreover, these scripts were designed for a specific cluster configuration available at the University of Waterloo.
The terminology in the paper and the repository/source code are not identical, this is as a consequence of the source code deriving from previous projects. Common name mappings are:
- site manager: data site
- site selector: adaptive storage advisor
- horizondb/Adapt-HTAP: Proteus