High-Performance DBMSs with io_uring: When and How to Use It

This repository contains the scripts, configurations, and plotting tooling required to reproduce all experiments reported in our paper.

Run experiments

# Install Nix on all servers.
ssh <node>
./scripts/install_nix.sh


cd experiments/

# Start Servers (runs in screen, shell can be closed)
./start_server.sh <node>


# Run experiments
nix-shell ../distexprunner
distexp-client -vv bench_nop.py

CSV files with measurements are stored in experiments/data/

Generate Plots

cd experiments/
nix-shell

for f in *.R; do
    Rscript $f;
done

PDFs are generated to experiments/out/

Manually run experiment:

mkdir build/
cd build/
cmake ..
make -j <binary>

Postgres

Create filesystem on SSD and mount it to /mnt/raid. Apply 0001-uring-opts.patch to postgresql/. Use script experiments/postgres/run_all.sh.

Hardware and OS Configuration

The experiments were executed on six identical cluster nodes with the following configuration:

CPU: AMD EPYC 9654P, 96 physical cores (192 hardware threads disabled, SMT off)
Memory: 768 GiB DDR5
SSDs: 8x KIOXIA KCMYXRUG7T68 (CM7-R Series) PCIe Gen 5 x4 (only on 1 machine)
NICs 400Gbit Mellanox Technologies MT2910 Family [ConnectX-7]

Operating System: Ubuntu Server 24.04 LTS
Kernel Version : Compiled from source: 6.15 and 6.17.0+
glibc Version  : 2.39
gcc Version    : 13.3.0
OFED:          : 24.10-1.1.4

Kernel installation

cd script/

# Use commit b988603f4f1e135f91789681fcbe076442cbc2c6 (with zero-copy recv bug-fixes)
REPO_URL=https://git.kernel.dk/linux.git BRANCH=for-next ./ubuntu_install_kernel.sh

# Select kernel for next boot:
./select_kernel.sh