Red Hat ブログ
Bare metal is making a comeback. At Red Hat we have been observing an increase of the use of bare metal in general. And we aren’t the only ones. In 2017’s OpenStack User Survey there had been a growth of bare metal in production environments from 9% to 20% of the production deployments. The 2018 survey says that adoption of Ironic is being driven by Kubernetes, with 37% of respondents who use Kubernetes on OpenStack using the bare metal provisioner.
And there are many reasons for this growth. A great blog post about Kubernetes on metal with OpenShift by Joe Fernandes described this growth in the context of containers on bare metal with Kubernetes as a driver for this growth. But, it doesn’t stop there - High-Performance Compute (HPC), access to hardware devices or scientific workloads such as AI/ML or data lake management are also contributing to this increase.
In order to provide safer and simpler access to what the Red Hat OpenStack Platform Team considers to be one of the best bare metal management platforms, we have been expanding our support for OpenStack Bare Metal (Ironic) for a number of releases now.
Ironic Support in Red Hat OpenStack Platform
At Red Hat we have been supporting Ironic with Red Hat OpenStack Platform (Red Hat OSP) for a while. Starting with Red Hat OSP 10 we added support to director, the OpenStack installer (known as TripleO upstream), to enable a highly available Ironic service in the control plane.
You can follow our official documentation to get it up and running.
OpenStack Bare Metal Certification Program
With OpenStack Platform 13 we introduced the Red Hat OpenStack Bare Metal Certification where server vendors may certify that their servers can be managed by OpenStack Bare Metal (Ironic).
OpenStack Bare Metal manages server nodes remotely via their baseboard management controllers (BMC) such as Dell’s iDrac, HPE’s iLO, Fujitsu’s iRMC, the Intelligent Platform Management Interface (IPMI), or Redfish, the DMTF open industry standard for remote management of servers via API in cloud environments, which has already been adopted by a number of vendors.
Red Hat’s new certification program is available to hardware vendors to have their servers tested for remote management by OpenStack Bare Metal (Ironic).
A More Efficient Way to Consume OpenStack Bare Metal-Managed Nodes
Along with this increase in interest in bare metal, we’ve been having several conversations with customers about providing a more simplified and cost effective way to manage large numbers of bare metal nodes with OpenStack.
They wanted to do this for a variety of use cases including:
One customer told us that they wanted a repeatable way to test the performance of every new revision of a hardware component for servers that they manufacture and Ironic was well-suited to include physical servers in a CI environment.
Another customer told us that they needed to perform 3D rendering in a large number of nodes and they had to automate the tests, so Ironic was their candidate.
We also see a growing number of customers who place their containers directly on the metal for performance, security and/or consolidation reasons.
Based on what we heard, we concluded that we had to make it easier to consume bare metal resources provided by Ironic. Starting in December 2018 we have also introduced a new SKU for Red Hat OpenStack Platform-managed bare metal nodes, which may help reduce costs in environments where OpenStack is managing bare metal.
OSP Director knows Bare Metal and OpenShift
Starting in Red Hat OSP 14, we added a feature to deploy Red Hat OpenShift Container Platform on bare metal directly from director. We have integrated hooks to invoke the openshift-ansible installer from director to set up Kubernetes and our container platform on servers. Since director uses Ironic, we leverage the features, the hardware ecosystem, and the experience that Ironic has accumulated over the years.
Red Hat OpenStack Platform director deploys OpenShift clusters on bare metal.
OSP director then, allows us to stand up an OpenShift Container Platform on bare metal, with nothing else than the OSP director server and the physical nodes we choose.
OpenStack Ironic to Manage Metal at Scale
Ironic is known for its ability to manage large amounts of servers distributed across racks. When the number of servers begin to scale out, more is needed than simply provisioning an operating system over the network to a bare metal node, you also need to think about:
Lifecycle management as part of day 2 operations for bare metal nodes,
Remote management via API
Command line or graphical user interface
Scaling out (or back in)
These capabilities and tools become increasingly important as these become part of an enterprise’s ability to innovate and differentiate their business. In this case, an OpenStack control plane (Overcloud) for bare metal can be deployed when high availability is desired.
An example of this is for customers deploying and maintaining OpenShift on OpenStack in a mixed environment with virtual and physical OpenStack instances. The architecture becomes a business critical operation as applications are developed and managed on OpenShift with OpenStack Ironic providing access to bare metal resources.
Ironic deployed by the OSP installer (director) and the openshift-ansible installer deploying OpenShift
The Roadmap with OpenStack Bare Metal
The certification of servers managed by OpenStack, the new subscriptions for the OpenStack-managed servers or the integration in the OpenStack installer to deploy OpenShift on bare metal are examples of our vision and commitment with OpenStack Bare Metal.
At Red Hat we’ve been striving to make OpenStack the best platform to manage bare metal nodes and the result is a platform with APIs and support for an ecosystem of servers and technologies.
What you can do today, as well as the roadmap, are equally exciting. New additions such as the Networking Ansible ML2 driver for multi-tenant BMaaS along with switch inspection via LLDP to verify VLAN or trunk configurations in switch ports, provisioning over routed networks supporting Spine and Leaf network topologies, Redfish integration or auto-discovery of nodes are examples of our innovation with OpenStack bare metal.