If you are a hyperscale (Cloud) data center manager, one of your top concerns is how to get the maximum amount of computing work done per Watt of power consumed. With that in concern at the forefront Cloud Providers like Google, Microsoft, and Facebook have strong incentives to explore new solutions for delivering compute cycles. Rumors coming out of Facebook suggest that it is looking to move away from its current X86 architecture platform in favor of servers based on ARM Holdings Cortex processor range. Porting an entire service to a new processor platform may not appear to be a sensible direction to take but porting to a new architecture is more a financial consideration than a technical one. If the cost per unit of performance justifies it , it is cheaper to pay a few programmers to rework the apps for a new architecture than it is to buy more servers.Dell clearly knows this, Paul Prince, (CTO Dell Enterprise Products Group) has been working on an ARM Cortex-A8 based server to complement Dell’s current range of Intel based semi-custom PowerEdge-C cloud boxes and now custom server vendor ZT Systems has announced a new server using low-power processors from STMicroelectronics based on ARM’s Cortex-A9 processor design. The new R1801e is a 1U rackmount server powered by up to 16 dual core SPEAr 1310 microprocessors. When fully populated the server has an impressively low power draw of less than 80 watts, more than low enough to catch the eye of any hyperscale data center owner.
ZT partnered with PHYTEC America to integrate the SPEAr 1310 with a server “System-on-Module” (SOM) architecture. Each SOM integrates the processor with 1 GB of DDR3 memory, 1 GB of NAND Flash, Ethernet and UART interfaces, presenting it as a single server module that supports a scalable architecture with up to eight discrete servers in each 1U enclosure. Two embedded switches connect the server modules together and provide uplinks over standard Ethernet, with system management for each SOM provided via industry standard Intelligent Platform Management Interface (IPMI). Interest in low-power processors has grown in recent years as data center operators struggle to manage rising power densities in their facilities, which often create cooling challenges. Much of the interest in ARM-based processors has focused on start-ups like Smooth-Stone ,which last week rebranded as Calxeda but has yet to launch a commercial product.
One possible disadvantage of ARM-based platforms is the current lack of availability of a commercial hypervisor solution to facilitate workload portability. Cloud service providers looking to increase the flexibility of ARM-based platforms will clearly benefit from the ability to leverage virtualization to simplify data center operations. Until then systems such as the R1801e will be limited to specialist applications. In the short-term this is unlikely to affect ZT systems business as its primary market is providing custom servers for large customers. Lack of support for virtualization limits the ability to concentrate server workloads onto the smallest number of physical server platforms, which in turn limits the power savings that can be made from powering down unnecessary servers. In the interim savings can be made through overclocking and under clocking processors in response to workload changes. However in the long term it is clear that hypervisor technology will be recall required to enable use of ARM-based processors to reach their full potential. ARM Holdings Has already acknowledged this requirement and is building direct support for virtualization into the next generation Cortex-A15 processor design. Processors based on the Cortex-A15 design will debut in 2012.
Projects to develop hypervisors for the ARM platform, have been under way for several years. The open source Xen community has an active project led by Samsung to develop a hypervisor for the Cortex-A9 processor (Samsung makes extensive use of ARM processor designs in its mobile phone business), although it is not clear if the Xen ARM hypervisor will be suitable for use on server platforms without further development. Other possible contenders in this market are Open Kernel Labs who has now deployed its ARM microvisor to over 1 billion phones worldwide, as well as UK startup B-labs that was founded by former ARM Holdings engineer Bahadir Balban. neither Citrix nor VMware have made public any plans to enter the ARM server market, although it is to be expected that both are watching developments closely and we should not forget that VMware has been showing off its mobile hypervisor platform for ARM processors since 2008.