If there were any notions that Moore’s Law still had some life to it and more powerful processors were on the way, several key developments over the past few weeks should put them to rest.
In the immediate future at least, greater computing power will come from harnessing multiple chips together, which means that in order to keep up with data trends, the enterprise will need to focus more on power consumption rather than power performance when evaluating new silicon platforms.
Most of the low-power action is coming from the ARM architecture. The technology already dominates smartphones and other mobile devices, so it only makes sense to employ the same technology in the data center. Indeed, ARMs are highly adept at moving the large-volume, small-packet traffic that is flooding the data center. So far, at least, multiple ARM processors have proven more adept in these environments than many-core x86 architectures.
But that might not be the case for much longer. Both AMD and Intel have committed themselves to developing low-power x86 devices for highly dense server configurations and new breeds of client devices. AMD recently released a pair of new quad-core Opterons, the X2150 and X1150, that operate on as little as 11 watts and are aimed at the new breed of microservers hitting the channel. The platform is based on the same Jaguar core design earmarked for upcoming Xbox and PlayStation 4 consoles, as well as HP’s new Moonshot platform.
Meanwhile, Intel is working on the next-generation “Haswell” Core architecture that is said to come in at less than 7 watts using the same 22 nm, 3D gate design that can be found on the Ivy Bridge processor. The Haswell design will most likely be used for PCs and “post-PC” devices like laptops, netbooks and hybrids. The chips feature native power management designed to match energy consumption with data loads and are said to be thinner and run cooler than standard x86 devices. And in a nod to the increased use of graphics processors for general purpose computing, the Haswell features an advanced GPU.
Increased reliance on large numbers of low-power processors also points to another facet of modern data architectures: the need for rapid scalability. The data center itself is becoming larger and more modular as massive web-facing operations like Google and Facebook seek to accommodate ever-increasing loads. Big, powerful servers like the HP ProLiant and the IBM Power lines are adept at handling massive databases, and will therefore still have a significant role to play in the era of Big Data, but the low-power architectures are much more flexible when it comes to the vagaries of web-based traffic. Indeed, with their ability to power up and power down at a moment’s notice, it almost seems that web-facing data centers themselves are starting to act like super-massive mainframe computers.
The low-power movement seems to have come along just in time to support top server vendors in what would otherwise be a miserable market. As IDC noted in its review of first quarter activity, Dell was one of the few to report positive revenues, largely by virtue of its low-power, high-density systems. As long as virtualization allows the enterprise to turn one large server into multiple smaller ones, it seems that demand for top-end machines will be light.
For the enterprise, low-power architectures are a blessing, but not a cure-all. Depending on the nature of your data loads and the legacy systems already in place, low-power systems can significantly boost performance and dramatically lower operating costs. Top-end systems, particularly virtualized ones, will still be preferable for traditional business applications, but new service-oriented, cloud-based environments will be better served by a low-cost, high-density architecture.