Install BEAST and BEAGLE on an AWS GPU instance

Terry Jones

2015-10-01 11:16

Quickstart
How I built the EC2 image
Install NVIDIA CUDA drivers
BEAGLE
Java 8
BEAST
- Testing BEAGLE
Make an EC2 image

Quickstart

If you just want to launch an AWS instance with BEAST installed along with BEAGLE with GPU support, use the EC2 AMI ami-6f63250a and launch either a g2.2xlarge (1 GPU, 15GB RAM, $0.65 per hour) or g2.8xlarge (4 GPUs, 60GB RAM, $2.60 per hour) instance running Ubuntu 14.04. Each GPU has 1,536 CUDA cores and 4GB of video memory.

Amazon EC2 instance types and pricing. Note that if you know you're going to be doing a lot of this, you will find it cheaper to purchase a reserved instance.

How I built the EC2 image

I started an AWS g2.2xlarge instance running Ubuntu 14.04. Then connect to it and get started:

$ ssh -i ~/dark-matter/aws/default.pem -l ubuntu 54.167.43.174  # Your IP address and keypair will be different!
$ sudo apt-get update
$ sudo apt-get upgrade

Install NVIDIA CUDA drivers

I followed instructions here with slight modifications to get a more recent version (7.5) of CUDA. The instructions on that page worked fine, but for the record here's what I did:

$ sudo apt-get update
$ sudo apt-get install build-essential
$ sudo apt-get install linux-image-extra-virtual

Take the "package maintainers version" when asked about the grub update.

Get CUDA:

$ curl -O -L 'http://developer.download.nvidia.com/compute/cuda/7.5/Prod/local_installers/cuda_7.5.18_linux.run'

This gets you version 7.5 of CUDA, which might be out of date by the time you read this. If you go to the NVIDIA CUDA download page and select your version of Linux etc., you can then copy the link of the Download button and use that instead of the 7.5.18 version mentioned above.

# Your version numbers may differ in the following.
$ chmod +x cuda_7.5.18_linux.run
$ mkdir nvidia_installers
$ sudo ./cuda_7.5.10_linux.run -a -extract=`pwd`/nvidia_installers
$ cd nvidia_installers
$ sudo ./NVIDIA-Linux-x86_64-352.39.run

Reboot:

$ sudo reboot

Log back into the instance.

cat <<EOT
blacklist nouveau
blacklist lbm-nouveau
options nouveau modeset=0
alias nouveau off
alias lbm-nouveau off
EOT | sudo tee -a /etc/modprobe.d/blacklist-nouveau.conf

Disable the Kernel Nouveau and reboot:

$ echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf
$ sudo update-initramfs -u
$ sudo reboot

Log back into the instance and install Linux source and headers. You can get your kernel version (and put it into the following) from uname -r (mine was 3.13.0-65, obviously).

$ sudo apt-get install linux-source
$ sudo apt-get install linux-headers-3.13.0-65-generic

Re-run NVIDIA installer:

$ cd nvidia_installers
$ sudo ./NVIDIA-Linux-x86_64-352.39.run -a

Load the kernel module:

$ sudo modprobe nvidia

Run the CUDA and samples installer:

$ sudo ./cuda-linux64-rel-7.5.18-19867135.run
$ sudo ./cuda-samples-linux-7.5.18-19867135.run

Verify CUDA is correctly installed

$ cd /usr/local/cuda/samples/1_Utilities/deviceQuery
$ sudo make
$ ./deviceQuery

You should see the following (your version number may differ) at the end of the output:

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 7.5, CUDA Runtime Version = 7.5, NumDevs = 1, Device0 = GRID K520
Result = PASS

BEAGLE

From the instructions on installing BEAGLE on Linux page:

$ sudo apt-get install autoconf automake libtool subversion pkg-config openjdk-6-jdk git openCL-dev

Then

$ git clone --depth=1 https://github.com/beagle-dev/beagle-lib.git
$ cd beagle-lib
$ ./autogen.sh
$ ./configure --prefix=/usr/local --with-opencl=/usr/lib/x86_64-linux-gnu
$ sudo make install

Java 8

From this page:

$ sudo add-apt-repository ppa:webupd8team/java
$ sudo apt-get update
$ sudo apt-get install oracle-java8-installer

BEAST

You might want to get a more recent version than the one below:

$ cd
$ curl -O -L 'https://github.com/CompEvol/beast2/releases/download/v2.3.1/BEAST.v2.3.1.Linux.tgz'
$ cd /usr/local
$ sudo tar xfz ~/BEAST.v2.3.1.Linux.tgz

See what BEAGLE resources BEAST thinks it has available

$ /usr/local/beast/bin/beast -beagle_info

If all has worked, the output of the last command above will look roughly like

BEAGLE resources available:
0 : CPU
    Flags: PRECISION_SINGLE PRECISION_DOUBLE COMPUTATION_SYNCH EIGEN_REAL EIGEN_COMPLEX SCALING_MANUAL SCALING_AUTO SCALING_ALWAYS SCALERS_RAW SCALERS_LOG VECTOR_SSE VECTOR_NONE THREADING_NONE PROCESSOR_CPU FRAMEWORK_CPU


1 : GRID K520
    Global memory (MB): 4096
    Clock speed (Ghz): 0.80
    Number of cores: 1536
    Flags: PRECISION_SINGLE PRECISION_DOUBLE COMPUTATION_SYNCH EIGEN_REAL EIGEN_COMPLEX SCALING_MANUAL SCALING_AUTO SCALING_ALWAYS SCALERS_RAW SCALERS_LOG VECTOR_NONE THREADING_NONE PROCESSOR_GPU FRAMEWORK_CUDA


2 : GRID K520 (OpenCL 1.2 CUDA)
    Global memory (MB): 4095
    Clock speed (Ghz): 0.80
    Number of multiprocessors: 8
    Flags: PRECISION_SINGLE PRECISION_DOUBLE COMPUTATION_SYNCH EIGEN_REAL EIGEN_COMPLEX SCALING_MANUAL SCALING_AUTO SCALING_ALWAYS SCALERS_RAW SCALERS_LOG VECTOR_NONE THREADING_NONE PROCESSOR_GPU FRAMEWORK_OPENCL

Testing BEAGLE

$ cd ~/beagle-lib
$ make check

Make an EC2 image

On the AWS control panel for EC2, select the instance you did the above on. In the "Actions" menu you can select Image / Create Image. As mentioned above, the image I created is ami-6f63250a.