The market for microservers – lower-powered, small form-factor system-on-a-chip (SoC) servers – is growing. Low-power servers became popular in 2011, but shipments for that year were a mere 19,000 units, just 0.2% of the total server market. In 2012, shipments rose to 88,000 units and reached 290,000 in 2013. Analysts at research firm IHS iSuppli predict that microservers will account for more than 10% of the server market by 2016.
Although microservers have yet to gain wide acceptance in the datacentre, demand for mobile, lightweight cloud and internet of things (IoT) workloads is driving growth, as Intel enters the market with its Xeon and Atom processor C2000.
Most servers are powered by x86 processors, but adoption of microservers, which could give datacentres simpler, denser and cheaper options, is gaining momentum. Rivals AMD and ARM have collaborated to capture the market, by replacing traditional power-guzzling x86-compatible processors with a 64-bit derivative of the ARM chip, though Intel is starting to provide low-power, high-density architecture.
Microservers can be grouped into clusters for tasks that do not require several CPUs. While they are not suited to every type of workload in a datacentre, they are becoming the preferred option for static web pages, basic content delivery and more. This is because a hosting environment supports computationally light tasks, such as requests for login authentication or serving small, static HTML pages. The large number of small nodes in microservers makes them ideal for such jobs.
Because microservers have a high density and energy-efficient design, their infrastructure (including the fan and power supply) can be shared by hundreds of physical server nodes, saving datacentre space.
More on microservers and micro datacentres
“Microservers provide a solution to the challenge of the increasing datacentre usage driven by mobile platforms,” say IHS iSuppli researchers. “With cloud computing and datacentres in high demand, specific aspects of server design, including maintenance, expandability, energy efficiency and low cost, are becoming more important.”
But they are not yet mainstream. Computationally intense tasks make up most jobs in an enterprise datacentre, which require traditional high performance server infrastructures based on the x86 platforms served by Intel. However, analyst firm Gartner estimates microservers will account for 4% to 5% of the revenue from the global server market by 2017. Experts say the fiercest battle will be between Cambridge-based ARM and x86 heavyweight Intel.
It is still early days for the adoption of microservers, says Chris Feltham, Intel’s European server manager: “We think current adoption is less than 1% but we are seeing interest and we take the architecture seriously. Microservers are definitely a big part of Intel’s strategy today as power costs are becoming a big priority.”
Intel’s Xeon and Atom chips
Intel provides processors for the smaller and larger ends of the microserver spectrum for enterprises’ lightweight workloads. The Intel Xeon processor E3 family offers a choice of node performance, performance per watt and flexibility, while the Intel Atom processor C2000 product family offers lower power and higher density.
The Xeon E3 1200 v3 chip features the highest single-threaded performance per node, better PCIe connectivity and graphic support, while the Atom processors C2000 and S1200 provide high-node density. The Atom processor C2000 product family is Intel’s second generation 8-core, 64-bit Atom SoC. It provides up to seven times higher performance, eight times more memory capacity and up to six times better power efficiency than the Atom S1200 series, according to Intel. The C2000, at 5 watts, is similar to Xeon E3’s 13 watts power consumption, but Xeon E3 family features x86-compatibility and includes datacentre-class features.
Even in the microserver category, there is no one-size-fits-all answer to system design and processor choice, says Feltham. Some microservers may have high-performing single-socket processors with robust memory and storage, while others have a far higher number of miniature dense configurations with lower power and low compute capacity per node. Intel pitches Xeon processor E3-based platforms for web-scale workloads requiring a highly dense infrastructure. The platform also boasts high performance demanding tasks in media, cloud or online gaming. Meanwhile, Atom C2000s are designed for lightweight scale-out workloads such as low-end front-end web and distributed memory caching.
“People think microservers are a solution to datacentre power problems, but that’s not true,” warns Feltham. “They also assume microservers are always cheaper, but that’s false too.”
Feltham says the cost of the processor should not be taken into account in isolation. The cost of management, the output derived from the processor and whether it meets the enterprise’s IT needs on time all together determine the total cost, he says.
Standardisation of Intel processors
But among the providers, Intel still controls this market, says Gartner’s analyst Errol Rasit. While microservers previously focused on low power, he expects Intel’s market presence will lead to high-performance microservers. Intel’s processor technology is considered more advanced compared with ARM’s architecture. “You get more peak performance per thread on Intel than on ARM-based chips,” says Nebojsa Novakovic, a Singapore-based consultant at Computational Resource Centre.
Intel is also investing in Haswell 22nm process and Broadwell 14nm process, both of which are designed to optimise the power savings and performance benefits. Intel’s Haswell line of chips are aimed at addressing enterprises’ low-power needs to power cloud and mobile lightweight workloads, while the Xeon processor E5 2600 v3 family, code-named Haswell EP, is based on DDR4, which cuts more power consumption according to Intel.
Rasit says that, because most datacentres have invested heavily in the x86 architecture, adding Intel’s own low-powered versions of microprocessors is easier for compatibility with the rest of the infrastructure, compared with an alternative chip such as ARM.
Every piece of innovation from competition has to be taken seriously
Errol Rasit, Gartner
“When you look at its competition, those architectures are based on an open-source ecosystem and there is no standard architecture yet, while the mainstream x86 server market is standardised around Intel,” says Rasit. “Standardisation, compatibility and the economies of scale give Intel an upper hand,” he says.
“There is a growing market for low-power CPU architecture and that’s where ARM is winning and that will be Intel’s biggest threat,” says US-based technology blogger Scott Wasson, speaking at an Intel workshop in August.
But according to Intel, standardisation can facilitate microserver innovation and it is working with companies, such as Tyan and Quanta, in the Server System Infrastructure (SSI) Forum to create the micro module server specification.
According to Intel’s Feltham, though, there is a lot of experimentation in the microserver market. “Some enterprises are testing and evaluating them but most CIOs are still interested in other IT priorities such as virtualisation and cloud computing,” he says.
But Rasit warns: “Intel shouldn’t be complacent because competitive architecture is very credible and giants such as Google will experiment with the alternative architectures powered by ARM.” Despite Intel having a stronghold in the server market, the equation could change if Intel takes its eyes off the ball, he says.
As power usage of datacentre products is increasingly becoming a priority for enterprises, microservers could become more than a device to fill the gap between PCs and servers, or a system for small and medium-sized enterprises that need a server but aren’t ready to invest in rack servers or blades, say experts.
“Intel must keep developing these architectures in parallel,” says Gartner’s Rasit. “Every piece of innovation from competition has to be taken seriously as there are some really credible and hungry competitors that are developing exceptionally good alternatives.”
The commoditisation of the standard X86-based server market will remain Intel’s bread and butter. However, Intel CIO, Diane Byrant, says that, with the launch of its Haswell and Avaton range of processors, designed for low-power computing, Intel may be moving from a “general-purpose compute provider” to delivering targeted systems for specific datacentre workloads. And Intel’s existing grip on the datacentre infrastructure could play a strong role in swaying users to adopt its own versions of low-power processors than alternative options.