Intel’s next goal is the datacentre rack

Intel can see an important change building in the way datacentres are used. With the growth in mobile oriented services and cloud-based service delivery in business operations, the company sees that the ability for datacentres to store, process and deliver streaming real-time services – be that video to a consumer mobile or consolidated global business accounts to the CEO - is fast becoming more important than runing or provisioning applications packages.

So the chip king has now declared a serious interest in addressing the next level of abstraction upwards in IT – the datacentre rack and its contents, as a package. The company is now looking to re-architect the compute resources needed in datacentres, and has already attracted the interest of leading US-based hosting services provider, Rackspace, which has announced it is to start following the reference racking architecture Intel is now proposing.

Given the company’s history of building functionality to its chip designs and its reference architectures, this can also be seen as the next putative `land-grab’ by Intel. In the same way that single-function chips such as processors, memories or graphics processors have merged into increasingly complex, multi-core System on Chip (SoC) devices, so it is possible to foretell that Intel is looking to absorb more and more of the functionality of a datacentre rack into the next generation of SoC units.  

Such a move would also map well onto the growing trend for the virtualisation of both datacentres and the networking infrastructure. This approach offers significant advances in the provision of service flexibility and agility by cloud service providers, and works best when there is a high level of commonality in the base architecture of datacentres, and high levels of commoditisation in the components used to build them.

By targeting the development of a new reference architecture, Intel can be seen to be trying to guide the provision of future compute resources in a direction favourable to its SoC capabilities.

To which the question must be: `and why not?’ As more mobile devices connect to the Internet, cloud-based software and applications are getting smarter by learning from the billions of people and machines using it, thus resulting in a new era of context-rich experiences and services. It also results in a massive amount of network connections and a continuous stream of real-time, unstructured data. As a result, datacentres must be more agile and service-driven than ever before, and easier to manage and operate.

This is what is underpinning Intel’s plans to virtualise the network, bring on new smart storage solutions and invest in innovative rack optimised architectures.

The company’s new Rack Scale Architecture (RSA), is an advanced design that promises to significantly increase the utilisation and flexibility of the datacentre to deliver new services.

Rackspace is likely to be just the first of many hosting companies that announce a start to the deployment of new server racks that mark the first step towards reaching Intel’s RSA vision. This, unsurprisingly, includes systems based on Intel Xeon processors and Ethernet controllers with storage accelerated by Intel Solid State Drives. The Rackspace design is the first commercial rack scale implementation.

The company sees the networking industry as now standing on the verge of a transition. Equipping the network with open, general purpose processing capabilities provides a way to maximise network bandwidth, significantly reduce cost and provide the flexibility to offer new services.

For example,the company estimates that with a virtualised, Software Defined Network (SDN), the time to provision a new service can be reduced to just minutes from two to three weeks with traditional networks. Intel has already introduced Open Network Platform reference designs to help OEMs build and deploy this new generation of networks.

Data growth is a challenge to all datacentres and transferring this large volume of data for processing within a traditional, rigid storage architecture is costly and time consuming. By implementing intelligent storage technologies and tools, Intel is helping to reduce the amount of data that needs to be stored, and is improving how data is used for new services.

Traditional servers are also evolving. To meet the diverse needs of datacentre operators who deploy everything from compute intensive database applications to consumer facing Web services that benefit from smaller, more energy-efficient processing, Intel outlined its plan to optimise workloads, including customised CPU and SoC configurations.

To this end the company has also announced a number of new processor options for now and the near future. Upcoming are two new version of the small, low-power Atom C2000 SoC family. One, codenamed “Avoton”,is aimed at the low-energy, high-density microserver and storage systems market, while the other, codenamed “Rangeley”, is aimed at network devices.

They will feature up to eight cores with integrated Ethernet and support for up to 64GB of memory, and are expected to deliver up to four times the energy efficiency and up to seven times  more performance than the first generation Intel Atom processor-based server SoCs introduced in December last year.

It has also announced plans for the next generation of Intel Atom processor SoCs (codenamed “Denverton”) that will enable even higher density deployments for datacentre operators.

Intel also disclosed an addition to its future roadmap – a new SoC designed from the ground up for the datacentre based on Intel’s next-generation Broadwell microarchitecture, which will form the heart of the next generation of Xeon E3 family processors. This SoC will offer higher levels of performance in high density, extreme energy efficient systems that datacentre operators will expect in an increasingly services-oriented, mobile environment.

Company executives, particularly on the parallel processing side of the business, have often talked about the notion of a hybrid multi-core processor module that would incorporate two Xeon processors for system management and specialised high compute demand applications, coupled with a large number of Atom processors to perform the majority of the work required. This is seen as an ideal configuration for a base level commodity module that can be used in both the typical datacentre and the more specialised high-performance and `supercomputer’ sectors.

All the signs are that Intel is now making a concerted move in that direction, as the company sees a strong move towards a far more service-provision operational model for all datacentres.