PaaS shoot-out: Cloud Foundry vs. OpenShift

OpenShift cartridges are pluggable components that can be combined within a single application. The built-in cartridges are different for the three versions of OpenShift, but the lists are all extensive, albeit limited to things that run on Red Hat Enterprise Linux (or Fedora, in the case of Origin, the open source version of OpenShift).

QuickStarts combine code and one or more cartridges, making it easy to install whole applications. While the OpenShift team doesn't maintain QuickStarts, anyone who is willing to be responsible for keeping one up-to-date with security issues is free to create and post one. WordPress, Drupal, and Ghost are among the top QuickStarts. Like Cloud Foundry buildpacks, OpenShift cartridges and QuickStarts are simple to build.

OpenShift runs applications in containers called gears. Cloud Foundry runs built and packaged applications called droplets, in Droplet Execution Agents. These containers are isolated from the other gears or droplets in the PaaS, and they are lightweight compared to virtual machines. In the future, both Cloud Foundry and OpenShift will support Docker containers.

Larger considerations
A key differentiator for Cloud Foundry -- that is, Pivotal's two Cloud Foundry offerings, Pivotal Web Services and Pivotal CF -- is support for the Pivotal Big Data Suite, which includes Pivotal HD (Pivotal's Hadoop distribution), HAWQ SQL for Hadoop, GemFire XD analytics, and the Spring for Apache Hadoop Java framework. The Pivotal Big Data Suite is an enterprise data warehouse that includes unlimited Pivotal HD.

According to Pivotal, in practice an administrator defines a service pool of HDFS and MapReduce instances, which take about five minutes to provision from scratch on Pivotal CF. Then a developer or an application can ask for an instance from the pool, obtain it in about two seconds, and a new instance can be created for the pool in the background. When the requested instance is no longer needed, it can be released.

Pivotal also offers a Mobile Services Suite that's integrated with both Pivotal CF and Pivotal HD. This is based on seven years and 400 apps' worth of know-how acquired with Xtreme Labs last year. It's basically an MBaaS (mobile back end as a service) on Pivotal's PaaS, with the integration extending out to mobile applications built on the platform.

A big differentiator for OpenShift is automatic application scaling, which adds gears and even nodes when an application becomes heavily used. It's built into OpenShift and doesn't require a front-end scaling service. You simply check a box when creating the application, then configure the traffic trigger points for adding and dropping gears.

Similarly, OpenShift automatically detects when an application is not getting any HTTP traffic and will eventually idle the gears without any intervention needed by the developer or operations. At such time as the application is requested again, OpenShift will automatically load it back into memory and handle the HTTP requests. OpenShift will even automatically start and restart misbehaving applications. All of these features reduce the amount of monitoring and operations work required to run an application on OpenShift.

Picking a PaaS
My experience installing OpenShift locally went very smoothly, once I got over my misguided attempt to use an unsupported virtual machine manager. My experience installing Cloud Foundry locally was not smooth, although ActiveState's Stackato variant was easier to install than Cloud Foundry open source. These experiences caused me to rate OpenShift higher than Cloud Foundry for installation. On the other hand, using the online version of each platform was a snap, and enterprises installing a PaaS in their own data center or on a private cloud are likely to use experienced consultants to do the setup.

The biggest pluses I see currently for OpenShift are its gear idling, which allows very high application density, and its automatic application scaling. These two features bring OpenShift up to a 9 on Management. Pivotal CF's lack of these features lower its Management score to 8. I think most enterprise customers will appreciate both gear idling and automatic application scaling. Note that Pivotal CF does have automatic application scaling on its road map.