17.2. Using AWS Instances

Many deep learning applications require significant amounts of computation. Your local machine might be too slow to solve these problems in a reasonable amount of time. Cloud computing services can give you access to more powerful computers to run the GPU intensive portions of this book. In this section, we will show you how to set up an instance. We will use Jupyter Notebooks to run code on AWS (Amazon Web Services). The walkthrough includes a number of steps:

  1. Request for a GPU instance.

  2. Optionally: install CUDA or use an AMI with CUDA preinstalled.

  3. Set up the corresponding MXNet GPU version.

This process applies to other instances (and other clouds), too, albeit with some minor modifications.

17.2.1. Registering Account and Logging In

First, we need to register an account at https://aws.amazon.com/. We strongly encourage you to use two-factor authentication for additional security. Furthermore, it is a good idea to set up detailed billing and spending alerts to avoid any unexpected surprises if you forget to suspend your computers. Note that you will need a credit card. After logging into your AWS account, click “EC2” (marked by the red box in Fig. 17.2.1) to go to the EC2 panel.

../_images/aws.png

Fig. 17.2.1 Open the EC2 console.

17.2.2. Creating and Running an EC2 Instance

Fig. 17.2.2 shows the EC2 panel with sensitive account information greyed out.

../_images/ec2.png

Fig. 17.2.2 EC2 panel.

17.2.2.1. Presetting Location

Select a nearby data center to reduce latency, e.g., “Oregon”. (marked by the red box in the top-right of Fig. 17.2.2) If you are located in China you can select a nearby Asia Pacific region, such as Seoul or Tokyo. Please note that some data centers may not have GPU instances.

17.2.2.2. Increasing Limits

Before choosing an instance, check if there are quantity restrictions by clicking the “Limits” label in the bar on the left as shown in Fig. 17.2.2. Fig. 17.2.3 shows an example of such a limitation. The account currently cannot open “p2.xlarge” instance per region. If you need to open one or more instances, click on the “Request limit increase” link to apply for a higher instance quota. Generally, it takes one business day to process an application.

../_images/limits.png

Fig. 17.2.3 Instance quantity restrictions.

17.2.2.3. Launching Instance

Next, click the “Launch Instance” button marked by the red box in Fig. 17.2.2 to launch your instance.

We begin by selecting a suitable AMI (AWS Machine Image). Enter “Ubuntu” in the search box (marked by the red box in Fig. 17.2.4):

../_images/ubuntu_new.png

Fig. 17.2.4 Choose an operating system.

EC2 provides many different instance configurations to choose from. This can sometimes feel overwhelming to a beginner. Here’s a table of suitable machines:

Name

GPU

Notes

g2

Grid K520

ancient

p2

Kepler K80

old but often cheap as spot

g3

Maxwell M60

good trade-off

p3

Volta V100

high performance for FP16

g4

Turing T4

inference optimized FP16/INT8

All the above servers come in multiple flavors indicating the number of GPUs used. For example, a p2.xlarge has 1 GPU and a p2.16xlarge has 16 GPUs and more memory. For more details see e.g., the AWS EC2 documentation or a summary page. For the purpose of illustration, a p2.xlarge will suffice (marked in red box of Fig. 17.2.5).

Note: you must use a GPU enabled instance with suitable drivers and a version of MXNet that is GPU enabled. Otherwise you will not see any benefit from using GPUs.

../_images/p2x.png

Fig. 17.2.5 Choose an instance.

So far, we have finished the first two of seven steps for launching an EC2 instance, as shown on the top of Fig. 17.2.6. In this example, we keep the default configurations for the steps “3. Configure Instance”, “5. Add Tags”, and “6. Configure Security Group”. Tap on “4. Add Storage” and increase the default hard disk size to 64 GB (marked in red box of Fig. 17.2.6). Note that CUDA by itself already takes up 4GB.

../_images/disk.png

Fig. 17.2.6 Modify instance hard disk size.

Finally, go to “7. Review” and click “Launch” to launch the configured instance. The system will now prompt you to select the key pair used to access the instance. If you do not have a key pair, select “Create a new key pair” in the first drop-down menu in Fig. 17.2.7 to generate a key pair. Subsequently, you can select “Choose an existing key pair” for this menu and then select the previously generated key pair. Click “Launch Instances” to launch the created instance.

../_images/keypair.png

Fig. 17.2.7 Select a key pair.

Make sure that you download the keypair and store it in a safe location if you generated a new one. This is your only way to SSH into the server. Click the instance ID shown in Fig. 17.2.8 to view the status of this instance.

../_images/launching.png

Fig. 17.2.8 Click the instance ID.

17.2.2.4. Connecting Instance

As shown in Fig. 17.2.9, after the instance state turns green, right-click the instance and select Connect to view the instance access method.

../_images/connect.png

Fig. 17.2.9 View instance access and startup method.

If this is a new key, it must not be publicly viewable for SSH to work. Go to the folder where you store D2L_key.pem (e.g., Downloads folder) and make the key to be not publicly viewable.

cd /Downloads  ## if D2L_key.pem is stored in Downloads folder
chmod 400 D2L_key.pem
../_images/chmod.png

Fig. 17.2.10 View instance access and startup method.

Now, copy the ssh command in the lower red box of Fig. 17.2.10 and paste onto the command line:

ssh -i "D2L_key.pem" ubuntu@ec2-xx-xxx-xxx-xxx.y.compute.amazonaws.com

When the command line prompts “Are you sure you want to continue connecting (yes/no)”, enter “yes” and press Enter to log into the instance.

Your server is ready now.

17.2.3. Installing CUDA

Before installing CUDA, be sure to update the instance with the latest drivers.

sudo apt-get update && sudo apt-get install -y build-essential git libgfortran3

Here we download CUDA 10.1. Visit NVIDIA’s official repository at (https://developer.nvidia.com/cuda-downloads) to find the download link of CUDA 10.1 as shown below.

../_images/cuda101.png

Fig. 17.2.11 Find the CUDA 10.1 download address.

Copy the instructions and paste them into the terminal to install CUDA 10.1.

## paste the copied link from CUDA website
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.243-418.87.00_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-1-local-10.1.243-418.87.00/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

After installing the program, run the following command to view the GPUs.

nvidia-smi

Finally, add CUDA to the library path to help other libraries find it.

echo "export LD_LIBRARY_PATH=\${LD_LIBRARY_PATH}:/usr/local/cuda/lib64" >> ~/.bashrc

17.2.4. Installing MXNet and Downloading the D2L Notebooks

First, to simplify the installation, you need to install Miniconda for Linux. The download link and file name are subject to changes, so please go the Miniconda website and click “Copy Link Address” as shown in Fig. 17.2.12.

../_images/miniconda.png

Fig. 17.2.12 Download Miniconda.

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
sh Miniconda3-latest-Linux-x86_64.sh

You need to answer the following questions:

Do you accept the license terms? [yes|no]
[no] >>> yes

Miniconda3 will now be installed into this location:
/home/ubuntu/miniconda3
  - Press ENTER to confirm the location
  - Press CTRL-C to abort the installation
  - Or specify a different location below
>>> <ENTER>

Do you wish the installer to initialize Miniconda3
by running conda init? [yes|no]
[no] >>> yes

After miniconda installation, run the following command to activate CUDA and Conda.

source ~/.bashrc

Next, download the code for this book.

sudo apt-get install unzip
mkdir d2l-en && cd d2l-en
wget https://d2l.ai/d2l-en.zip
unzip d2l-en.zip && rm d2l-en.zip

Then create the conda d2l environment and enter y to proceed with the installation.

conda create --name d2l

## Package Plan ##

  environment location: /home/ubuntu/.conda/envs/d2l

Proceed ([y]/n)? y

After creating the d2l environment, activate it and install pip.

conda activate d2l
conda install pip

Finally, install MXNet and d2l. The postfix cu101mkl means that this is the CUDA 10.1 variant with Intel MKL support. For different versions, say only CUDA 10.0, you would want to choose cu100 instead.

## mxnet
pip install mxnet-cu101mkl --pre
pip install git+https://github.com/d2l-ai/d2l-en@numpy2

You can test quickly whether everything went well as follows:

$ python
>>> from mxnet import np, npx
>>> np.zeros((1024, 1024), ctx=npx.gpu())

17.2.5. Running Jupyter

To run Jupyter remotely you need to use SSH port forwarding. After all, the server in the cloud does not have a monitor or keyboard. For this, log into your server from your desktop (or laptop) as follows.

# This command must be run in the local command line
ssh -i "/path/to/key.pem" ubuntu@ec2-xx-xxx-xxx-xxx.y.compute.amazonaws.com -L 8889:localhost:8888
conda activate d2l
jupyter notebook

Fig. 17.2.13 shows the possible output after you run Jupyter Notebook. The last row is the URL for port 8888.

../_images/jupyter.png

Fig. 17.2.13 Output after running Jupyter Notebook. The last row is the URL for port 8888.

Since you used port forwarding to port 8889 you will need to replace the port number and use the secret as given by Jupyter when opening the URL in your local browser.

17.2.6. Closing Unused Instances

As cloud services are billed by the time of use, you should close instances that are not being used. Note that there are alternatives: “Stopping” an instance means that you will be able to start it again. This is akin to switching off the power for your regular server. However, stopped instances will still be billed a small amount for the hard disk space retained. “Terminate” deletes all data associated with it. This includes the disk, hence you cannot start it again. Only do this if you know that you will not need it in the future.

If you want to use the instance as a template for many more instances, right-click on the example in Figure 14.16 Fig. 17.2.9 and select “Image” \(\rightarrow\) “Create” to create an image of the instance. Once this is complete, select “Instance State” \(\rightarrow\) “Terminate” to terminate the instance. The next time you want to use this instance, you can follow the steps for creating and running an EC2 instance described in this section to create an instance based on the saved image. The only difference is that, in “1. Choose AMI” shown in Fig. 17.2.4, you must use the “My AMIs” option on the left to select your saved image. The created instance will retain the information stored on the image hard disk. For example, you will not have to reinstall CUDA and other runtime environments.

17.2.7. Summary

  • Cloud computing services offer a wide variety of GPU servers.

  • You can launch and stop instances on demand without having to buy and build your own computer.

  • You need to install suitable GPU drivers before you can use them.

17.2.8. Exercises

  1. The cloud offers convenience, but it does not come cheap. Find out how to launch spot instances to see how to reduce prices.

  2. Experiment with different GPU servers. How fast are they?

  3. Experiment with multi-GPU servers. How well can you scale things up?

17.2.9. Discussions

image0