MATLAB on AWS Stack/Instance

Hi,

I've been successful running MATLAB on AWS using info from the MATLAB provided github link, but I have several of questions related to the creation of a stack via the provided info:

1. The instructional MATLAB video shows how to go through all the steps, but the "Cleaning up" instructions at the end just deletes the stack, and I assume that this would also delete any data that I've uploaded to the virtual machine. How can I stop the stack/instance and keep my data for later use?
2. Separate from the github link, I've seen that I can create an instance and select the amount of RAM, storage and desired GPU. It looks like the github link only permits me to select storage and GPU.....is there a way to also select the amount of RAM via the stack creation method?
3. When setting up an instance, I found that the Amazon Machine Image lists a number of R202Xa/R202Xb windows instances in the Community AMIs....will these help me resove #1 and/or #2?

I'm a novice to AWS services, but I'd like to leverage the GPUs that are available and try to get this working for a number of big projects - I'm hoping that someone can provide some good info.

Thanks for your help!

Answer 1

[David Willingham]

Hi Matt,

Firstly, if you'd like to automate the creation of the VM on AWS, I'd recommend using our Deep Learning Container on DockerHub or from NGC. It will help automate setting up the VM if you're creating and destroying it based on your need for more compute on AWS.

There are instructions on how to use containers using docker cam be found in this blog post. NOTE: The 2nd half of the blog is related to NGC, it can also apply to docker hub too.

To answer your questions:

1. If you want to save data in between instances you create or destory:

• Save any data to a cloud drive, such as s3. Here is a doc page on transferring to s3.
• Setup source control, such as Git, and store your files in GitHub. Information on setting up souce control directly in MATLAB can be found here. If you following this programming practise, you can "pull" the files you need from GitHub after setting up a new instance.

2. I normally select the AWS instance type that suits the needs I have for the models I'm training at the time. AWS's p3 instances are recommended if you require a GPU, and theres 4 different options.

3. In the blog post above, it highlights how you can select the NVIDIA AMI.

David

[Matt K.]

Thanks for the reply. I'll be sure to check out all the links!