Deployment with AWS Elastic Container Service

Justin Coyne

Stanford Libraries

Samvera Connect - October 2018

An abbreviated history of cloud computing

cc by-nc-nd by Aristocrat https://www.flickr.com/photos/netphotography/15036969665

In the beginning there was the server room

cc by-nc-nd by Berkley Lab https://www.flickr.com/photos/berkeleylab/4157700219

Then Jeff Bezos offered to rent you a virtual machine (2006)

cc by Steve Jurvetson https://www.flickr.com/photos/jurvetson/5129286606

and it was good.

CC by-nc loren chipman https://www.flickr.com/photos/97902275@N05/41767706762

Containers appeared (2008)

Container as a Service (CaaS)

Consistent environment

cc by-nc-nd by Adam Vogel https://www.flickr.com/photos/newt0/6614704109

Isolation

cc by-nc by David Ingram https://www.flickr.com/photos/dingatx/8219989409

Run Anywhere

cc by-nc-nd by Mina Guli https://www.flickr.com/photos/run4water/33318217996

Better deployment

Amazon Elastic Container Service (ECS)

At Stanford we've been using ECS for over a year now and we're pretty happy with out decision. I'm going to walk through what we do to spin this up. ECS has two modes, Fargate launch type and EC2 launch type. With Fargate launch type, all you have to do is package your application in containers, specify the CPU and memory requirements, define networking and IAM policies, and launch the application. EC2 launch type allows you to have server-level, more granular control over the infrastructure that runs your container applications. With EC2 launch type, you can use Amazon ECS to manage a cluster of ec2 instances and schedule placement of containers on the servers. You are responsible for provisioning, patching, and scaling clusters of ec2 instances. We've found that for our needs Fargate is far easier to set up and doesn't have instances that we have to patch. It allows us to focus on delivering value to the user.

AWS Terms

Cluster - e.g.: The production environment
Task - Which containers should I run and how many resources
Service - Network, Load balance, Auto scale

Cost

The price per vCPU is $0.00001406 per second ($36.43 per month) and per GB memory is $0.00000353 per second ($9.14 per month).

~$600 per year

You can turn it off when you're not using it!

Set up autoscaling to add capacity when you need it and scale back when you don't.