Skip to content

INSTALLATION.md

Latest commit

 

History

History
122 lines (98 loc) · 3.46 KB

INSTALLATION.md

File metadata and controls

122 lines (98 loc) · 3.46 KB
Raw

Installation

This guide details the installation and basic usage of the GPU Query Optimizer in Postgres.

Building postgres

Prerequisites

  1. Install CUDA following the instructions: https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html

  2. verify that CUDA is installed and functional:

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Mar_21_19:15:46_PDT_2021
Cuda compilation tools, release 11.3, V11.3.58
Build cuda_11.3.r11.3/compiler.29745058_0
  1. find CUDA installation path (in my case /usr/local/cuda):
$ which nvcc
/usr/local/cuda/bin/nvcc
  1. find GPU Compute Capability
$ /usr/local/cuda/extras/demo_suite/deviceQuery | grep Capability
  CUDA Capability Major/Minor version number:    6.1

Installation

  1. from the build folder (which can be the same as the repository root), run the configure script
# (optional) enable bitmapset extension if available
export CFLAGS="-march=native -mtune=native"
export CPPFLAGS="-march=native -mtune=native"

# where do you want to install
PG_PREFIX=/home/user/postgres/opt
mkdir -p $PG_PREFIX

# assuming you want to build in-tree, otherwise use full path to configure
# --with-icu enables unicode (used by some databases we experimented on)
# we also added --without-readline because readline was not available on our machine
./configure --prefix=$PG_PREFIX \
    --with-icu \
    --enable-cuda=/usr/local/cuda \
    --with-cudasm=61

# for debugging you might want to enable the following options:
# --enable-debug
# --enable-cassert
# --enable-depend
# and also disable compiler optimizations with
# export CFLAGS="-O0"
  1. from the build folder run make
# see README for list of available options.
# we always built with profiling enabled which prints additional timing
# information.
# For debugging purposes, you might want to compile with enable_debug=yes
# pass -jN for faster build time
make -j $(nproc) enable_gpuqo_profiling=yes

  1. finally, run make install

  2. (optional) run make -j $(nproc) world and make install-world to get the extensions (used in Musicbrainz database)

Setup

  1. add the Postgres installation path to your PATH
export PATH=$PG_PREFIX/bin:$PATH
  1. create new PGDATA folder
#
export PGDATA=/home/user/postgres/data
mkdir -p $PGDATA

initdb
  1. run the just built Postgres daemon
PORT=6543 # default postgres port is 5432

setsid postgres -p $PORT > /tmp/pgout 2> /tmp/pgerr &

# check that it is running
less /tmp/pgerr
# LOG:  database system is ready to accept connections

# to stop the daemon you can run
# kill $(head -n 1 < $PGDATA/postmaster.pid)

Other Build Options

Timing information can be printed to stdout for gpuqo algorithms setting enable_gpuqo_profiling=yes in make:

make enable_gpuqo_profiling=yes

Other useful make variables:

  • enable_debug=[yes/no]: enable debugging symbols generation
  • cost_function=[cout/simple/postgres]: choose cost function used by gpuqo algorithms: cout (C_out), simple (join-order-benchmark paper, with the addition of sort-merge joins), postgres (returns the same cost as Postgres but only supports a limited set of queries)
  • simulate_complex_cost_function=[yes/no]: add overhead to the cost function
  • disable_ccc=[yes/no]: globally disables CCC (warp divergence avoidance method). The gpuqo_dpsub algorithm can be run without ccc at runtime (no need to set this option).