Intel Eyes Integrated Fabric Controller in Xeons for Cloud, Big Data

Intel in the next few years will integrate a fabric controller into its Xeon server chips, which promise better performance and efficiency in cloud, HPC and dense environments.

Intel, which already houses the memory controller and graphics capabilities on its processors, within the next few years will also integrate the compute fabric controller on its server chips, a move that officials say will speed up performance, improve scalability and enhance energy efficiency in data centers.

An integrated converged fabric controller will eventually be included in the chip maker's Xeon server processors, according to Raj Hazra, vice president of the company's Intel Architecture Group and general manager of technical computing at Intel. However, Hazra would not say when the integration would occur.

The move will come as enterprises increase their adoption cloud computing and virtualization, as big data takes an increasingly big role in business and as Web-based companies like Google, Facebook, Twitter and Amazon look for ways to grow the performance and scalability of their massive data centers while looking to drive down the costs of powering and cooling them.

Data center fabrics essentially enable enterprises to more easily scale their data center compute capabilities, and more closely link storage and networking technologies, Hazra told eWEEK a few days before the kickoff of the chip vendor's Intel Developer Forum 2012 show in San Francisco, which officially starts Sept. 12.

A fabric helps businesses address the growing needs for lower latency, greater bandwidth, improved compute density and low power in data centers by connecting the various systems and components, from chips and memory to servers, storage and network, according to Hazra. Having a converged fabric controller integrated on the processor will enable fabrics to more efficiently and intelligently move data around the data center while allowing for smaller, more dense and more power-efficient servers.

It also will help fabrics better direct data around data centers that can incorporate multiple interconnect technologies, from Ethernet to InfiniBand to PCI-Express (PCIe). And different computing environments require different jobs from their fabrics, he said, from high-performance computing (HPC) clusters-which demand greater bandwidth, power, scalability and message rates-and public clouds (bandwidth and scalability) to enterprise applications (bandwidth) and microservers (density, power and scalabilty).

"There's no one fabric, and in many cases, as with HPC and cloud, it's often more than just one," Hazra said.

An integrated fabric controller-currently fabric controllers are found outside the processor-will result in fewer components in the server node itself, reduced power consumption by getting rid of the system I/O interface and greater efficiency and performance. According to Hazra, the bandwidth along the PCIe system I/O interface between the processor runs and fabric controller runs at 32GB per second; the bandwidth from the fabric controller into the fabric runs at 10 to 20G bps. With an integrated fabric controller, the system I/O interface not only is eliminated, but the controller will offer a bandwidth of more than 100G bps, he said.

Intel already has much of its fabric story in place, he said, from its Xeon chips and Xeon Phi coprocessors-as well as its low-power Atom platform-to various data management solutions, HPC tools, and Ethernet, InfiniBand and HPC interconnect technologies. In addition, Intel over the past couple of years has been aggressive in acquiring other technologies that will factor into its fabric efforts, from its acquisition in 2011 of Fulcrum Microsystems for its Ethernet solutions to the $125 million it paid for QLogic's InfiniBand IP and the $140 million for supercomputer maker Cray's HPC interconnect solutions this year.

Intel is not the only chip maker pursuing integrated fabric capabilities. Rival Advanced Micro Devices in February bought microserver maker SeaMicro for $334 million. SeaMicro had been selling systems based on Intel's Atom chips. A key to AMD's purchase was the ability to get SeaMicro's Freedom Fabric technology, which handles storage and networking virtualization.

Charles King, principal analyst with Pund-IT Research, said in a research note at the time of the acquisition that SeaMicro's Freedom Fabric "is really the jewel in the crown of the AMD deal. Not only should the Freedom technology bolster AMD's considerable efforts in high performance and supercomputing, but the company's OEM customers that are focused on those and related cloud and Web 2.0 markets (which is to say, most all of them) will likely consider the technology a valuable addition to their solution quivers."

For their part, Intel executives said they considered buying SeaMicro, but eventually passed, noting that they had their own microserver and fabric efforts under way.