Gartner Inc. recently brought the gaping functional gaps between containers and conventional virtualization into the spotlight with a widely-covered report that concluded that the average organization is still better off using a hypervisor than a container for most workloads.

That isn’t stopping some web-scale companies from forging ahead with containers at scale, however. CIOs should take note of their experiences, since practically every major new enterprise trend of the past few years has started as a born-on-the-web innovation.

The most notable adopter of containers is Google, which uses an internal implementation to deliver its services to billions of users every week. However, a growing number of startups are achieving results with publicly available alternatives.

A glimpse of the future

One of the most prevalent advantages cited by adopters of container-powered production environments is the ability to deploy applications as multiple independent modules, which makes it possible to update the individual components faster and more reliably. That’s how Gilt Groupe Holdings, Inc.’s Gilt.com online fashion retailer designed its services.

The model exploits the fact that containers typically use considerably less hardware resources than virtual machines, which means that a new instance can come online in seconds rather than minutes. That allows administrators to avoid the hassle of updating live machines and instead simply replace an image with an entirely new copy, which is not only simpler but also provides the option to quickly revert back to the original in the event of a problem.

The ability to roll out changes so rapidly makes it possible for developers to make changes much more quickly than in a typical enterprise setting. In a recent keynote speech, Gilt founder Michael Bryzek claimed that his startup implements about 100 updates every day while keeping downtime to a few hours per year.

But unlocking the full potential of containers requires a fair amount of work. For Shopify Inc., another early adopter, the starting point was making instances as lightweight as possible to minimize the time it takes to fire up new images. After an experimentation period that engineer Graeme Johnson described as “less painful than it sounds,” the e-commerce company settled on a base configuration incorporating only the roughly two dozen components that every type of container in its environment requires to launch.

Shopify also pushed the time-consuming process of compiling the different assets, a task normally performed during the deployment of new instances, upstream to the build phase while stripping off all the other elements from the standard image in order to minimize duplication. That included the crucial configuration management component for handling log aggregation and analytics, which it moved into a dedicated container on each host.

Both Johnson and Gilt’s Bryzek stressed the importance of monitoring, pointing out that providing administrators with proper visibility into the environment is the prerequisite to ensuring reliability. However, that’s easier said than done due to the stateless nature of containers, which required Shopify to create a custom daemon to collect the data coming off its instances.

The e-commerce startup and Gilt have proven that running containers in production is already well within the realm of possibility, although the amount of time and effort required to pull off a large-scale deployment may make the technology impractical for the typical enterprise. But as the paradigm slowly moves towards becoming an enterprise reality, CIOs need to start drawing lessons from the experience of their web-scale peers and establish a course toward the service-driven architectures of tomorrow.

photo:Cole Burston