Big Data

Verizon puts 50,000 Docker containers to work on a massive IoT cluster

IoT is becoming real, but it demands serious infrastructure, as a peek inside Verizon's newest cloud shows.

Cloud

Launching a Docker container is so yesterday. Today, it's all about 50,000 Docker containers... launched in 100 seconds... on a Mesos-powered cluster.

This is exactly what Verizon did today at MesosCon in Seattle.

Some think "Can you hear me now?" when they think of Verizon, but in the cloudy world of infrastructure, the telecom giant boasts its share of technical achievements. Indeed, Verizon may now have one of the world's fastest clusters for Internet of Things (IoT) services.

At MesosCon in Seattle, the company demonstrated 50,000 Docker containers launched in 100 seconds, on a Verizon cluster powered by the Mesosphere Datacenter Operating System (DCOS). It's all tied to the surge in IoT business Verizon anticipates.

Verizon's State of the Market for The Internet of Things 2015 cited 45% year-over-year revenue growth in 2014 for its IoT business, with 4G LTE activations growing by 135%. With more than 15 million—and growing—IoT-enabled connections under management at Verizon, the company is aggressively beefing up its back end to support the expected growth. As I've written, IoT is driving huge data center investments, but most of that growth is happening in clouds like Verizon's.

I spoke recently with Larry Rau, director of Technology at Verizon, to learn more.

TechRepublic: How has IoT affected Verizon's development and operations?

Rau: We're really seeing an explosion of IoT devices and data, and that's generating a huge surge in demand for new IoT services. The business opportunities around IoT have pushed us to look at ways to accelerate the velocity our developers can build and deploy those services, while meeting the performance guarantees Verizon provides its customers.

For us, that's meant re-configuring our IoT stack to support what we internally call a philosophy of "build today and deploy tomorrow."

The traditional ways that large enterprises deploy new applications are tied up in a lot of process. You have to claim resources in the datacenter, you have to stand up hardware, you have to apply additional human operators, consider power issues, and have teams involved for networking and datacenter management.

Old approaches to vertical scale and manually partitioned hardware created cycles that could drag out deployments of new services for months. Moving to a cluster-oriented solution allows Verizon to aggregate datacenter resources across all services, enable automation, and deliver faster time to market for new services.

On the data side, the volume and variety of data generated by IoT devices has created a number of new scale considerations.

Verizon doesn't have a challenge with the network part of moving data—we run the fastest network around—but our re-configuring addresses how we reliably store that data and make it available to IoT services we are creating around real-time, post-processing, and analytics. A cluster that supports the ability to run a mix of application types allows different forms of processing to occur on the same data, often concurrent with each other.

TechRepublic: From what I've seen, Verizon is one of the largest users of Mesosphere's DCOS. What's the role of the datacenter operating system in your environment?

Rau: Large scale clustering will be the centerpiece of our new IoT platform. Mesosphere takes a much more holistic view of the cluster and allows us to go beyond the limitations of old vertical scale approaches and the pain of associated high human-touch synchronization.

We have a very large cluster available to our IoT services. It's all abstracted as a single machine and allows us to launch containers at scale without the human configuration and capital expenses that used to go into claiming and standing up hardware.

Mesosphere allows traditional manual administration to be automated and enables our developers the ability to accelerate the speed at which new services can be deployed.

DCOS will enable Verizon to condense deployment times by an order of magnitude. Where it could take the average business months to deploy a new service for millions of users of Verizon's services, DCOS will enable that cycle to be reduced up to 90%.

Verizon doesn't do anything small. When we deploy a new product or service, it has to be geographically dispersed, and it has to have five nine's of reliability. DCOS allows us to just tap into the capacity of our existing cluster and introduce new services that reliably scale from day one.

TechRepublic: In its vision for the IoT business opportunity, Verizon has been outspoken about the importance of how data is handled in the cloud. How will the new data surge in IoT affect the typical business' back end?

Rau: The data gathered by millions of connected endpoints must be stored reliably and securely and made accessible to people and other systems. As your IoT requirements grow, your infrastructure needs to support that growth.

As I said, we don't have a challenge with the network part of moving data. However, when you get data into the cloud, then the challenge becomes how to take different types of actions with that data.

One of the big advantages of the cluster abstractions and working with DCOS is handling different types of processes and actions without having to redundantly move that data around. This allows us to lower I/O operations by moving tasks around without having to move the data—and that scenario applies to real-time, post-processing, or analytics after the fact.

We also use DCOS to allow related frameworks like Kafka and Cassandra to run on the same cluster, so we gain the benefits of sharing the resources underneath.

Also see

About Matt Asay

Matt Asay is a veteran technology columnist who has written for CNET, ReadWrite, and other tech media. Asay has also held a variety of executive roles with leading mobile and big data software companies.

Editor's Picks

Free Newsletters, In your Inbox