ARM-based processing will without a doubt make its way into the enterprise soon. But this is not as dramatic a change as it might seem, as it has more to do with accommodating disparate user devices than remaking long-standing server infrastructure.
The ARM, of course, is not only a low-power solution compared to the x86 devices that populate most server farms, it also happens to be the processor of choice for the vast majority of mobile devices that are seeking access to enterprise networks. So it makes perfect sense for server infrastructure to utilize the same basic technology found in client devices.
This certainly isn't lost on the likes of Dell, which just took the wraps off the new ARM-based Copper server while at the same time opening up the Crowbar management framework. And in a cagey bit of marketing, the company will ship the new servers to select customers for testing - kind of like a "Tricks are for Kids!" approach that builds brand loyalty by placing top customers in an exclusive club while sowing envy among everyone else.
The thing about ARMs, however, is that they are low power in multiple senses. True, they consume less wattage that a typical x86, but they also deliver less processing chip for chip. Not to worry, say ARM backers, because you can pack more cores into a given footprint to match processing requirements and still undercut the x86 power envelope. All true, but only if standard enterprise applications can be multithreaded to such a high degree, which, as we've seen in the struggles of multicore parallelism, is proving to be rather more difficult than it sounds.
This is part of the reason ARM Holdings has backed efforts like CARP (Correct and Efficient Accelerator Programming), a European initiative to develop advanced programming tools based on the OpenCL format. Although aimed at mobile applications, the growing interest of ARM as an enterprise solution should result in a fair amount of crossover. And since OpenCL is designed to support highly parallel environments consisting of GPU and other acceleration products, the group's goal of bridging high-level DSL code to low-level OpenCL APIs through a "portable intermediate language" (PIL) should make it easier for developers to build multithreaded apps for ARM environments.
Multithreading is only a concern when it comes to traditional enterprise applications, however. As Web-based commerce and cloud computing evolve, the need for large numbers of low-power processing points is likely to increase. Chip designers like Calxeda are targeting advanced ARM designs around Web-serving, content delivery and Big Data applications, which prize scalability and tighter resource utilization over raw power. In this new world, which the company is addressing through its EnergyCore processor, applications that are I/O- or memory-intensive are favored over those that require top CPU horsepower.
And that's the true key to understanding the advent of ARMs in the enterprise. Far from taking over, ARMs are uniquely qualified to handle much of the new data and application loads that will be generated by mobile and cloud-based environments. That will likely be a substantial portion of the overall data environment, but it isn't everything.
The challenge will be to foster these disparate architectures without building new data silos.