Installation¶

Operating Systems¶

• CentOS/RHEL 7.0 or later
• Ubuntu 15.04 or later

Additional operating systems might be available on a case-by-case basis.

Libraries¶

• libjvm, provided by Java 1.6 or later
• libldap

GPU Cards (Hardware)¶

MapD currently supports NVIDIA Kepler and Pascal series GPU cards.

Drivers (GPU Installations Only)¶

• NVIDIA GPU Drivers version 375.51 or later.

If you do not use those drivers, you must install with GLX/Xorg. Contact MapD support for more information.

CentOS / Red Hat Enterprise Linux (RHEL)¶

libjvm is provided by the packages java-1.7.0-openjdk-headless or java-1.8.0-openjdk-headless. 1.8.0 is preferred, but not currently required. To install run:

sudo yum install java-1.8.0-openjdk-headless

By default a symlink pointing to the newly installed JRE is placed at /usr/lib/jvm/jre-1.8.0-openjdk, with the libjvm library residing in the subdirectory lib/amd64/server. Add this subdirectory to your LD_LIBRARY_PATH environment variable in order to start the MapD Core Server. Use the following command:

export LD_LIBRARY_PATH=/usr/lib/jvm/jre-1.8.0-openjdk/lib/amd64/server:$LD_LIBRARY_PATH

You can add this command to any file managing your environment, such as $HOME/.bash_profile, /etc/profile, or /etc/profile.d/java.sh.

Ubuntu / Debian¶

libjvm is provided by the package default-jre-headless. To install, run:

sudo apt install default-jre-headless

By default, a symlink pointing to the newly installed JRE is placed at /usr/lib/jvm/default-java, with the libjvm library residing in the subdirectory jre/lib/amd64/server. Add this subdirectory to your LD_LIBRARY_PATH environment variable in order to start the MapD Core Server. Use the following command:

export LD_LIBRARY_PATH=/usr/lib/jvm/default-java/jre/lib/amd64/server:$LD_LIBRARY_PATH

You can add this command to any file managing your environment, such as $HOME/.bash_profile, /etc/profile, or /etc/profile.d/java.sh.

CentOS / Red Hat Enterprise Linux (RHEL)¶

RHEL-based distributions require Dynamic Kernel Module Support (DKMS) in order to build the GPU driver kernel modules, which is provided by the Extra Packages for Enterprise Linux (EPEL) repository. See the EPEL website for complete instructions for enabling this repository.

1. Enable EPEL

   sudo yum install epel-release
   

2. Update and reboot The GPU drivers and their dependencies, specifically kernel-devel and kernel-headers, require that the latest available kernel is installed and active. To ensure this, update the entire system and reboot to activate the latest kernel:

   sudo yum update
   sudo reboot
   

3. Download the NVIDIA CUDA Toolkit network install RPM package, but do not follow the installation instructions on the NVIDIA site (the instructions are in the next step). Download the RPM package from NVIDIA CUDA Zone.

   Note: The “local” NVIDIA driver RPM contains the older drivers, which do not include the fix required for EGL on CentOS. The “network” RPM provides the updated drivers. In general, you should use the network RPM, as it always provides the latest drivers. The downside of using the network install is that it requires Internet access. If yum update pulls in new drivers, you must reboot your machine.

4. Use the following commands to install the CUDA repository, update local repository cache, and then install the GPU drivers. yum automatically installs additional dependencies for the cuda-drivers package, including dkms, gcc, kernel-devel, and kernel-headers. The CUDA Toolkit (package cuda) is not required to run MapD Core, but the GPU drivers (package cuda-drivers, which include libcuda) are.

   sudo rpm --install cuda-repo-rhel7-8.0.44-1.x86_64.rpm
   sudo yum clean expire-cache
   sudo yum install cuda-drivers
   

   Where cuda-repo-rhel7-8.0.44-1.x86_64.rpm is the name of the RPM package provided by NVIDIA.

5. Reboot and continue to section Environment Variables below.

Ubuntu / Debian¶

Download the DEB package provided by NVIDIA from the NVIDIA CUDA Zone.

1. Install the CUDA repository, update local repository cache, and then install the CUDA Toolkit and GPU drivers

   sudo dpkg --install cuda-repo-ubuntu1604_8.0.44-1_amd64.deb
   sudo apt update
   sudo apt install cuda-drivers linux-image-extra-virtual
   

   Where cuda-repo-ubuntu1604_8.0.44-1_amd64.deb is the name of the package provided by NVIDIA.

2. Reboot and continue to section Environment Variables below.

Environment Variables¶

For CPU-only installations of MapD Core, skip to section MapD Installation below.

MapD Core Server depends on libcuda, which must be available in your environment in order to run MapD Core. The NVIDIA GPU drivers usually make libcuda available by default by installing it to a system-wide lib directory such as /usr/lib64 (on CentOS/RHEL) or /usr/lib/x86_64-linux-gnu (on Ubuntu).

Verifying Installation¶

After installing CUDA and setting up the environment variables, restart your machine to activate the GPU drivers.

On Linux, you can verify installation of the GPU drivers by running nvidia-smi.