Young-Sae Song is a VP at AMD
Virtualization, perhaps more than any other technology, is powering tremendous growth in cloud-based services. At its heart, virtualization is a hypervisor that allows the provisioning of virtual machines on a single piece of hardware.
For cloud services the hypervisor is the key enabler for two reasons.
First it allows the underlying hardware to be divided up, allowing some organizations to maximize their hardware utilization, and second, it provides a way of provisioning new services without having to be in the data room. The economic benefits of increasing hardware utilization are obvious; however, it is a single shot of efficiency. But the ability to provision services on demand is the feature which fuels the continual growth in demand for cloud services.
On-demand provisioning, which provides users with an elastic infrastructure to meet the varying usage demands, is possible due to a combination of the hypervisor and supporting services that are accessed by application programming interfaces (APIs). Through the use of APIs, third-party applications give users the ability to control virtual machines as if they were independent pieces of hardware.
But what if performance or regulatory requirements call for single-tenant, dedicated hardware?
That’s where bare-metal provisioning comes into play with OpenStack’s Ironic project set to bring the flexibility of virtual-machine provisioning to physical servers.
Why provision bare metal?
Cloud service providers such as SoftLayer, which was recently acquired by IBM, have demonstrated an appreciable appetite for single-tenant hardware in the datacenter. Having dedicated hardware offers organizations an extra layer of data security and dependability, ensuring that compute and storage resources are not being shared with others. And, equally as important, having dedicated hardware offers the ability to know exactly what performance is available, meaning service providers can offer improved, higher specification service-level agreements (SLA).
Applications such as relational database servers require considerable compute and storage resources, up to a point where using virtual machines is no longer a viable option to deliver acceptable performance and economies of scale. While the need for virtual machines may have gone away, the need to meet fluctuating usage patterns hasn’t. This means that system administrations still need the ability to spin up servers as if they were virtual machines with SLAs in place to ensure dependable performance.
Large, monolithic software packages are not the only reason why bare-metal provisioning makes sense. Organizations that handle sensitive data may feel more comfortable being able to guarantee that customers’ data is being stored on hardware that is only serving a single tenant.
For service providers, bare-metal products allow for higher performance SLAs without concerns that other virtual machines may lead to server-wide performance degradation. Bare-metal provisioning provides a win-win situation for customers looking for increased performance and security benefits afforded by dedicated hardware, and service providers who can offer premium SLAs.
Getting down to the metal
Over the past decade, a movement started to get closer to the metal, to remove the layers of software abstraction between the operating system and the hardware. Early hypervisors ran on top of operating systems and were soon superseded by bare-metal hypervisors, which incorporated a lightweight operating system and the hypervisor in one distribution, with the goal of increasing overall system performance.
Even with improvements in bare-metal hypervisors, a performance overhead still exists, and for some applications that performance hit is an unacceptable outcome. But as previously stated, hypervisors and their associated APIs provide more than just the ability to divide hardware, they also offer a means of provisioning new services using minimal effort.
The OpenStack project has been working on bare-metal provisioning through the Ironic project. It will bring provisioning flexibility seen with virtual machines to physical servers, allowing service providers to integrate bare-metal provisioning within existing virtual machine management systems with minimal changes to the software infrastructure.
While the Ironic project has been working on getting bare-metal provisioning ready at the software level, AMD has been testing bare-metal provisioning in its SeaMicro SM15000 server, ensuring that customers who need the benefits of bare metal can provision and manage a bare-metal server deployment as easily as they do virtualized servers. The SeaMicro SM15000 provides the ability to boot and install server images from a central server and offers APIs that can be integrated into OpenStack environments.
Bare-metal provisioning through OpenStack’s Ironic project will give service providers the ability to offer physical servers as if they were virtual machines and open up new possibilities for organizations and service providers.
Young-Sae Song is Corporate Vice President of Product Marketing at Data Center Server Solutions for AMD. In this role, he leads the outbound marketing, branding, and demand generation functions for AMD’s push into next generation fabric based computing systems. Song brings more than 18 years of experience marketing innovative enterprise software and networking products.