The 17th release of the open-source OpenStack cloud platform is now generally available, with new capabilities to support advanced workloads with improved container integration.
The OpenStack Queens platform was officially released on Feb. 28, marking the 17th release of the open-source cloud platform, originally started by NASA and Rackspace in 2010. OpenStack today is widely used by large organizations, including Walmart, as well as serving as the underlying infrastructure for multiple cloud providers, including platforms from IBM and Oracle, among others.
Multiple new and enhanced capabilities have landed in the OpenStack Queens release, including virtual GPU (vGPU) support and improved container integration. Several new projects also have made an appearance in the OpenStack Queens milestone, including Cyborg, which provides a framework for managing hardware and software acceleration resources.
“The biggest thing we’re seeing in the market now is that people want to do more with their cloud deployments,” Mark Collier, chief operating officer of the OpenStack Foundation, told eWEEK. “There is a proliferation of new workloads like machine learning, artificial intelligence and containers, and there is a lot in the Queens’ release for those use cases.”
The OpenStack Queens release follows the Pike release, which was launched in August 2017. Queens is different from Pike in multiple ways, not the least of which is the fact that Queens has more new features.
“Pike had few new things, but it had a higher percentage of work that was oriented around operations,” Jonathan Bryce, executive director of the OpenStack Foundation, told eWEEK.
Among the new features in OpenStack Queens is vGPU support. Bryce explained that the vGPU support is in the OpenStack Nova compute project.
“The vGPU capability builds on support that Nova has been working on for the last several releases with the placement service,” he said.
The Nova placement service is a dynamic inventory system within OpenStack that keeps track of what resources are available for deployment, according to Bryce. The vGPU capabilities are hardware-specific and work with the Nvidia Grid vGPU and Intel GVT (Graphics Virtualization Technology).
“A GPU might have thousands of cores, and if you just try to do a generic passthrough of the hardware into the virtual machine, what you end up with is the whole GPU inside of the virtual machine,” Bryce said. “What a lot of companies want is a fleet of different GPU servers that can be divided up among different workloads, and up until now that wasn’t something that was easily managed in a multitenant way.”
The new Cyborg project comes from the telecom world. It provides a generic framework for acceleration of all kinds and is not GPU-specific. Bryce said Cyborg can make use of physical GPUs, FPGAs (Field Programmable Gate Arrays) or other kinds of accelerators.
OpenStack also already had the Ironic Bare Metal service that enables cloud operators to directly access physical hardware assets. Cyborg is complementary to what Ironic provides, Bryce said.
“They are both tools to manage hardware directly,” he said. “Ironic is oriented around managing whole servers, while Cyborg takes a hardware function view for acceleration.”
Another new initiative that has landed in the Queens release is the OpenStack Helm project. Helm is a popular open-source project that works as a package manager for the Kubernetes container orchestration system.
OpenStack Helm joins multiple other efforts, including OpenStack-Ansible, which aims to provide some form of configuration and deployment management capabilities for OpenStack services.
“Both OpenStack Helm and OpenStack Ansible are similar in that they are tools to help manage OpenStack services, but they take different approaches,” Bryce said.
Helm uses Kubernetes for its lifecycle management approach and is a container-native approach. Bryce noted that some organizations are more comfortable using a configuration management tool, which is what Ansible provides.
The Lightweight Open Container Initiative (LOCI) project is another new effort that is joining the OpenStack Queens platform. The OCI is a Cloud Native Computing Foundation (CNCF) specification to create an industry standard for container runtimes. LOCI provides an alternative to the existing OpenStack Kola project, which Bryce said has a more complete packaging approach for each container image.
“LOCI takes an approach that is more aligned with the Kubernetes way of running an image, where the container itself is very small and the management sits outside of the container,” Bryce said.
The Rocky Road Ahead
With OpenStack Queens now out, development is starting on the next major release, which will be known as OpenStack Rocky. Although it’s still early in the development cycle, there are a few items that are likely to be included in the OpenStack Rocky release.
One of such feature Fast Forward Upgrade, which will enable cloud operators to more easily upgrade to OpenStack releases. Currently, operators need to move from one OpenStack release to the next successor release. For example, an OpenStack Pike platform can move to Queens, but a Pike platform could not directly update to Rocky when it becomes available.
“With Fast Forward Upgrades, operators can skip ahead more than one release,” Collier said. “Upgrades have always been a pain point for users as not everyone relishes the prospect of upgrading their cloud platform every six months.”