Virtualization Technology News and Information
VMblog's Expert Interviews: Part 1 - Diamanti Talks Managing Kubernetes and Docker Containers. Concerns and Pitfalls of DIY


Diamanti, a San Jose-based startup I last interviewed in November 2017, develops an appliance to help enterprises stand up a Kubernetes cluster running Docker containers in minutes. CEO Jeff Chou is a big believer in the benefits of Kubernetes and Docker containers on bare metal. He thinks VMs are already for legacy apps and a poor choice for cloud-native workloads going forward. In this two-part QA series, I explore the issues Chou raises in more detail. There is no doubt that container adoption is exploding. Gartner predicts that "by 2018 more than 50% of new workloads will be deployed into containers." But how do you want to run your containers? That is becoming a very difficult question for IT organizations everywhere. Let's hear from Diamanti.

VMblog:  Where are most enterprises today in their thinking on the right infrastructure for running containers?

Jeff Chou:  Existing IT infrastructure has been optimized over the past decade for virtualized business applications. But VMs may not support containers the way you want to run them. You may need new infrastructure. Given the increasing pressure on IT teams, you'll also want to consider carefully whether you will buy or build your infrastructure stack to support containers. As with any infrastructure decision, there are five main factors to consider:

  1. Solution components. What components-hardware and software-make up the solution, and how well do they meet your requirements?
  2. Deployment time. How long will it take? How much expertise will that require? Solutions that exceed your team's skills require expensive professional services engagements, adding cost and time.
  3. Management. How much time and expertise will the various hardware and software components require to manage?
  4. Scalability. How difficult is it to scale the solution as your needs grow?
  5. Total cost of ownership (TCO). How much will it cost you to own and operate the solution, including staff time?

VMblog:  What are the key actors you need to understand for container environments?

Chou:  I see five main factors a team needs to think through carefully before they start any container journey.

  1. Bare-metal or virtualized? Bare metal is the gold standard for production containers. Running containers inside VMs adds an additional layer to the stack that must be managed and debugged, adding cost and complexity.
  2. Persistent storage. Applications running inside containers need to be able to save data permanently. Container solutions need to provide a mechanism for persistent storage, even as containers come and go.
  3. Networking model. Getting networking right remains one of the most difficult aspects of container environments, and container networking must integrate smoothly with your existing datacenter networking.
  4. Orchestration. Dynamic container environments require orchestration tools to coordinate activities and automate operations. While there are many options, Kubernetes has emerged as the clear leader.
  5. Support. How will you get support for the full infrastructure stack, including both hardware and software? From a single vendor? From multiple vendors? From the open-source community?

VMblog:  Tell me more about the particular challenges of DIY versus buying a container solution off-the-shelf, and what is the specific impact on IT infrastructure for storage and networking that should be considered first?

Chou:  There is no easy way to solve this problem for all organizations. But beware the urge to DIY. We saw this movie before with PaaS. In the end, pretty much every Fortune 500 organization dropped its DIY efforts in favor of a commercially-supported solution like OpenShift or Cloud Foundry. We're about to repeat those DIY mistakes with containers and Kubernetes.

Just as we saw with PaaS, if you choose to build out infrastructure yourself, deployment will take longer and will almost certainly require professional services to complete. Ongoing management complexity will be higher than it would be with a more integrated solution since you'll have to keep up with the open-source community and integrate patches and enhancements as needed.

If you opt for servers with internal storage, scaling is straightforward. You simply add another server when resources run low, but you'll also have to provide a mechanism for data management and data protection on each server, figure out how to balance storage use across the set of servers, and you may need to provide a mechanism for shared storage such as NFS or a clustered file system such as Ceph or GlusterFS.

If you choose separate storage, it can simplify storage management initially, but scaling becomes more complicated. Suppose you start with a few servers and one storage array. To scale, you add servers until the storage array runs out of performance or capacity. Then you either add a second storage system or replace the original system with a more powerful one. Either way, this can be disruptive and can result in a big, and often unanticipated, incremental expense.

When it comes to networking, you're completely on your own. You need to make sure that your chosen networking model is compatible with your container and orchestration solutions. Again, that may not sound difficult, but the networking documentation for technologies like Docker and Kubernetes run to hundreds of pages, suggesting there's a lot to think about and plan for.


In the second part of our interview, we'll take a deeper look at alternatives to DIY with their respective pros and cons. Chou says there are some appealing solutions from large enterprise vendors, but they're also complex and customers are forced to pay the VMware licensing tax. But one solution gives you the benefits of bare metal to run Kubernetes clusters and Docker containers and is almost literally plug-and-play. Stay tuned.
Published Tuesday, January 16, 2018 9:35 AM by David Marshall
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
<January 2018>