README.md

HotSpots

This is the repo for our databases class project. Our report is in report.pdf.

We modify the concurrent B-tree with OLC from https://github.com/wangziqi2016/index-microbench.

Building

> git clone https://github.com/mark-i-m/hot-spots
> cd hot-spots
> git submodule update --init --recursive
> make

You need to have g++ 7 or newer, along with GNU make.

To run tests for B-tree implementations: make tst.all. To run a test foo: make foo.tst.

To build the report: make report.pdf.

Running workloads

All benchmarks were run using the benchmarks/run_eval.sh script in this repo. It allows varying the number of insertion and lookup threads, the B-tree implementation, and the number of operations per thread, the initial load of the B-tree, and the frequency with which to output statistics. See the usage message for more info:

> cd benchmarks
> bash run_eval.sh # prints help message

To run benchmarks, you need to have passwordless sudo access. You also need to have cpupower installed. To get this tool, install apt install linux-tools-common. Then, attempt to run cpupower. This will fail with a message that tells you what else to install (it is kernel-specific). Install the first package only.

Plotting graphs

The results of the workloads are produced in the results directory. To graph results, use the results/plot.py script. The figures in the figures directory were generated using this script. See the usage message for more info:

> ./plot.py # prints the help message

This script requires python 3, numpy, and matplotlib.

Organization

The organization of this repository is as follows:

• README.md: This README.
• makefile: The makefile for building and running everything.
• benchmarks/: Implementation of benchmarks and utilities for running.
• btrees/: Implementation of B-trees.
• figures/: Figures produced from results with the plot.py script.
• results/: Results produced from the run_eval.sh script.
• tests/: Correctness tests for the B-trees.
• report.tex, references.bib: Our report.