New Features in Queens
HPC/AI/Machine Learning
Support for vGPUs (virtual graphic processing units) — In Nova, vGPU support lets cloud administrators define flavors to request specific resources and resolutions for vGPUs. End users can boot VMs that have vGPUs, an important capacity for graphics-intensive workloads and many scientific, artificial intelligence and machine learning workloads.
High Availability
Cyborg — Cyborg, a framework for managing hardware and software acceleration resources such as GPU, FPGA, CryptoCards and DPDK/SPDK, debuts in the Queens release. Acceleration has become a necessity rather than an option, particularly for telcos with NFV workloads. With Cyborg, operators can list, identify and discover accelerators, attach and detach accelerators to an instance, and install and uninstall drives. It can be used standalone or in conjunction with Nova or Ironic.
Enterprise Workloads
Cinder Multi-Attach enables operators to attach the same Cinder volume to multiple VMs. If one node goes down, the other takes over and has access to the volume. This redundancy—which supports high availability (HA) for mission critical workloads—is one of the most-requested features in cloud environments and has remained a largely unmet challenge in computing until now.
Edge Computing
OpenStack-Helm — This addition to the project portfolio provides a collection of Helm charts and tools for managing the lifecycle of OpenStack on top of Kubernetes and running OpenStack projects as independent services.
Containers
LOCI — Another project making its debut, LOCI makes Open Container Initiative-compatible images of OpenStack services that can be dropped into heavy-weight deployment tools like OpenStack-Helm or used individually to deliver standalone services like Cinder block storage.
Kuryr CNI Daemon — OpenStack is the preferred platform for containers deployed in private cloud, and the community continues to expand microservices features in Queens. Kuryr adds a CNI daemon to increase the scalability of operations on Kubernetes. To support HA, the CNI daemon watches for pod events, eliminating the need to wait on the Kubernetes API for each event. Pods can be created even if the controller goes down.
Operator Enhancements
Ironic Rescue Mode — Instance repair—long available for VMs in Nova—is now available for bare metal instances in Ironic. Operators can now troubleshoot misconfigured bare metal nodes or recover from issues like a lost SSH key, an important enhancement since production usage of Ironic has jumped from 9 to 20 percent between April and November of 2017 (source: OpenStack User Survey).
Register & document policy — Across the majority of OpenStack projects, role-based access control (RBAC) policies now live in the project code, as opposed to being a separate file in the project source, providing better communication about service policies and the ability to set more granular defaults for RBAC policies.
To see more release highlights from the project teams, visit the Release Highlights page. Learn more about individual OpenStack Projects in the Project Navigator, and get the full view of OpenStack with the Project Map.