Last week, I highlighted the growing divergence between traditional data centers that exist throughout the business world and the new breed of web-facing infrastructure manned by Google, Facebook and the like.
The point was that both hardware and software development seem to be following different paths depending on whether the environments they support cater to the high-volume but relatively basic nature of Web transactions or the more complex requirements of business and scientific applications.
Nowhere is this dichotomy playing out more openly than in the basic unit of enterprise architecture: the processor.
Much has been made of the advent of ARM processors in enterprise-class servers, with many observers saying this represents the beginning of the end for x86 platforms, at least as far as the typical enterprise environment is concerned. Not only is the ARM already driving the fortunes of most mobile platforms, but it provides an extremely low-power solution just as enterprises are feeling the heat, so to speak, to lessen the power draw in the data center.
All true. And yet, I can’t help but notice that the modern enterprise is tasked with providing optimal solutions, not just energy-efficient ones. And there are still many applications that require high power from a single processor, or at best a small number of cores, rather than banks of low-power devices.
Research taking place at the Federal Polytechnic Institute’s (EPFL) EcoCloud research center in Switzerland is starting to bear this out. The group notes that large-scale replacement of current x86 architectures with low-power units will provide considerable cost-savings, but primarily for Web-facing environments that deal mainly with simple memory requests. Once the workloads shift over to more complex processing, the benefits decrease because productivity per core starts to fall off.
This is a crucial point because, as The Linley Group noted recently, the enterprise industry is quickly shifting its processor evaluation criteria from “performance per processor” to “performance per watt.” With both cloud computing and high-performance computing now the leading drivers of server sales, the time is ripe for new designs that capitalize on the needs of these two market segments. For some, that means greater emphasis on ARM-equipped microservers and blades, while others will likely gravitate toward higher-end designs.
This partly explains some of the latest moves by arch-rivals AMD and Intel. AMD made waves last month with the announcement that it will integrate ARM’s 64-bit Cortex-A50 architecture into the Opteron family starting in 2014. The company is planning on an eight-core device to start, most likely using the SeaMicro Freedom Fabric as a means to court integration into leading microserver designs. At the same time, Intel has doubled down on the Itanium processor with the new 9500 series. The company is betting that it will gain increasing converts from RISC architectures even as it courts x86 users through the Common Platform program designed to increase commonality with its large installed base of Xeons.
It seems likely, then, that we are headed toward a parallel universe in the enterprise. On one hand, there is the traditional data environment of databases and applications; on the other is the new world of Web traffic and mobile interactivity.
Two data centers, two processor architectures and two types of workload.