July 20–24, 2015
Portland, OR

Container orchestration at scale

Steve Francia (Google)
9:00am–12:30pm Tuesday, 07/21/2015
Scale Portland 252
Average rating: ****.
(4.53, 17 ratings)

Prerequisite Knowledge

No prior knowledge of orchestration is required. However, the audience should be familiar with at least one container runtime, like LXC or Docker, to get the most out of this session.

Materials or downloads needed in advance

We will create cloud virtual machines with Docker pre-installed and preconfigured for this training. It allows us to jump straight into the action, without bothering about installing Docker in the possibly very diverse environments of the audience. It also reduces the strain on conference internet access, because instead of downloading VM and container images, the audience interacts with the training servers through low-bandwidth SSH connections. Therefore, all you need is a computer with an SSH client installed.


All applications are different, and deployment cannot be solved by a one-size-fits-all product. We want to cover many different scales, from modest web applications to giant behemoths spanning thousands of machines.

But first, what exactly is this thing we call “container orchestration?”

We are probably all familiar with the basic resource management concern that arises when allocating virtual machines on a compute cluster. It is an easy task when we run dozens or even hundreds of virtual machines on a handful of physical nodes. However, it becomes exponentially complex when dealing with millions of virtual machines on thousands of computers, spanning multiple datacenters, with high availability and latency requirements translating into placement constraints. How do we programmatically find out where to start the 50 virtual machines required by this new project? More importantly, how do we ensure that we make the most efficient possible use of available resources? This is orchestration.

Not everybody has to deal with that kind of scale. However, with containers and microservice architectures becoming increasingly pervasive, smaller applications are often broken down into dozens of components. Long before reaching hundreds of nodes, we need an orchestration solution to deploy and manage those components across our infrastructure.

In this workshop, we will show multiple container orchestration products in action. We will also expose their pros and cons, and show how to integrate with network plumbing to achieve load balancing and high availability. When walking out of this lab, participants will no longer think that orchestration is a complex topic.

Photo of Steve Francia

Steve Francia


Steve Francia is a Gopher at Google as well as an author, speaker, and developer. Steve is the creator of Hugo, Cobra, spf13-vim. Previously, he was an exec at Docker and MongoDB. Steve serves on the board of Drupal.

Comments on this page are now closed.


Picture of Jerome Petazzoni
Jerome Petazzoni
07/20/2015 12:48pm PDT

@Dave: I’ve been able to do this tutorial to smaller “test” audiences, and I found that presenting PAAS products was too much material in a half-day tutorial.

Regarding the requirements: each person attending the tutorial will be provided with credentials allowing to access their own cluster of VMs. The connection to those VMs is made over SSH. That’s why you need a SSH client.


Picture of Dave McAllister
Dave McAllister
07/13/2015 4:23am PDT

On the tutorial requirements. The email asks for a SSH client. Any reason that Terminal isn’t enough?

Picture of Dave McAllister
Dave McAllister
07/13/2015 4:22am PDT

I think that we should be careful in promoting too many possible solutions. For example, I’d pitch opendeis as my choice of Docker PaaS.

Picture of Magno Torres
Magno Torres
06/22/2015 7:55am PDT

That’s great! And if you want to know more about tsuru, feel free to contact us anytime

Picture of Jerome Petazzoni
Jerome Petazzoni
05/30/2015 5:49am PDT

I plan to cover (for sure) Compose, Swarm; very probably Mesos, some Kubernetes; maybe mention Dokku and other PAASes. That being said, if a dozen people indicate here that they would love to see a Docker-based PAAS in action, I can do that and thin out the other sections a bit, for sure. The main challenge is the time constraint :-)

Picture of Magno Torres
Magno Torres
05/26/2015 9:00am PDT

Will you cover tsuru PaaS? I’m PO of this project, I and my team can help any doubts