Hosting firm Rackspace is introducing a new service that combine the on-demand nature and scalability of cloud servers with the performance and total control of bare-metal servers. This is aimed at meeting the requirements of specific workloads such as databases or Hadoop deployments.
Known as OnMetal Cloud Servers, the new service will be available from July, initially from Rackspace’s Northern Virginia data centre only, but is expected to roll out internationally during 2015.
The service brings all the power and flexibility of cloud computing to applications previously considered unsuitable to run in a virtualised environment, according to the firm. It is an API-driven, single-tenant infrastructure-as-a-service (IaaS) offering that enables customers to provision dedicated servers with whatever operating system and services stack they require.
Rackspace has been looking at bare-metal provisioning since at least last year, when the firm introduced its Performance Cloud Servers tier for customers with more demanding workloads. However, there has been growing interest in the ability to own the entire server, according to the firm, because of the “noisy neighbour” problem in multi-tenant environments, where another workload on the same host may degrade network latency, disk input/output (I/O) and compute processing power.
Rackspace president Taylor Rhodes said: “Virtualisation and sharing a physical machine are fantastic tools for specific workloads at certain scale; however, we’ve learned that the one-size-fits-all approach to multi-tenancy just doesn’t work once you become successful, so we created OnMetal to simplify scaling for customers to stay lean and fast with a laser-sharp focus on building out their product.”
OnMetal Cloud Servers make use of the Ironic Bare Metal Provisioning project in the OpenStack cloud computing framework. This is still in incubation rather than a full core part of OpenStack, but Rackspace has a policy of introducing cutting-edge features in its cloud services.
The physical hardware itself is compliant with Open Compute Project specifications, and available in three different tiers aimed at specific workloads.
These comprise a compute-optimised configuration for application servers supporting 20 threads and 32GB memory, while a memory-optimised configuration for tasks such as in-memory analytics supports 24 threads and 512GB.
An I/O-optimised configuration supports 40 threads with 128GB memory and a 3.2TB PCI Express flash drive. The latter is best for traditional databases, NoSQL and online transaction-processing applications, Rackspace said.
Pricing has not been disclosed, but Rackspace said customers will be able to pay by the minute, with utility-style billing only for the resources they use.
Rackspace is not the first provider to offer such a service. UK-based Bigstep has been offering its Full Metal Cloud service since last year.