Intel to Launch Super-Virtualized, Fabric-Based Datacenter

A glimpse of Intel’s future?

Intel has a new CEO, a new president, and, once it’s officially announced July 22, a new long-term strategy for products aimed at datacenter installations.

New Intel CEO Brian Krzanich and President Renee James have said they want to expand Intel’s presence in markets such as mobile and wearable computing.

Analysts placed much of the blame for the 25 percent drop in Intel’s earnings during the first quarter of this year on the ongoing decline in the PC market and Intel’s failure to make headway in tablets and smartphones.

Intel demonstrated a mobile chip code-named Medfield in January 2012 as part of an effort to expand its presence in mobile, and has gotten its silicon into products such as Samsung’s Galaxy Tab 3. Overall, however, chips from ARM Holdings still dominate the smartphone and tablet markets.

Fresh opportunities such as wearable connected computing devices, meanwhile, will take time to grow enough to make an impact on Intel’s revenues, according to Krzanich, who is optimistic about the opportunity anyway. “I think it will take about two years to roll that plan out, but I think you’ll see an explosion,” he told CNN, speaking about both the growth potential of Google Glass and Intel’s plan to help build out the ecosystem of processors supporting wearable computing, rather than just supplying chips.

Intel already has a formidable presence within the datacenter and plans to expand that presence even further, according to Diane Bryant, senior VP and general manager of Intel’s Datacenter and Connected Systems Group, who has been evangelizing the plan at international conferences and media interviews since at least April.

At the April Intel Developer’s Forum in Beijing, Bryant called Intel’s datacenter products the broadest of any chipmaker, but said Intel’s plans go far beyond its current strength in powering servers with x86-based Xeon processors and its power-conserving Intel Atom system-on-a-chip product line.

The basic strategy underlying Bryant’s “Reimagine the Datacenter” banner rests on expanding the abilities of Intel chips to provide better-integrated, higher-performing versions designed to maximize CPU performance, memory, storage and networking in rack-mounted systems. The goal is to use high-bandwidth interconnects such as Intel’s Silicon Photonics to “hyperscale” Intel-based datacenter systems and virtualize control of individual servers to make their resource-distribution more efficient.

The design of rack-mounted server systems, according to Intel materials supporting Bryant’s April 10 talk, will change as the systems are forced to scale up, first by changing physical configurations to make them easier to access and cool, and later with virtualization complete enough to let datacenter managers assign CPU cycles, memory and other resources of a single server to different workloads. Processing, memory and I/O will become modular subsystems attached to servers rather than indivisible parts of a single server.

Modular computing components connected using high-speed fabric interconnects and virtualized storage services, in theory, should give datacenter managers much more granular control over what resources to assign particular workloads. They will also maximize utilization by finding underused memory or I/O and reassigning the idle portion of their capacity, Bryant said.

Intel is building a series of products and integrated sets of products designed to virtualize, manage and scale datacenter computing resources in just that way, combining the ability of Intel’s Xeon and Atom SoC processors with all the other functions of the server, much as Cisco has tried to do with its unified-computing systems.

The latency of the interconnect Intel is developing is so low latency the modules themselves can be located in places other than a particular spot on the motherboard. That effort is part of a broader push by vendors to catch the growing market for microservers, interconnects and low-power processors connected via high-speed interconnect into densely packed clusters, along with server farms designed for hyperscale implementations.

Those high-performance, low-requirement setups are possible only if microservers and racks can become far more densely packed, easier to cool and maintain, gulp far less energy and cost little enough that they are able to grow toward exascale without becoming unmanageable on any of those fronts, according to Peter Lin, analyst for compute platforms at analyst firm IHS.

Demand for high-density, high-efficiency datacenter hardware is growing so fast that the percentage of datacenter hardware that includes microservers will jump from 0.2 percent in 2011 to more than 10 percent in 2016, according to an IHS study published in April.

The details of the plan Intel will announce later this month are still unclear, but it’s very likely both the strategy and products included in it will resemble Project Scorpio—a rack-scale architecture Intel is developing in conjunction with China Telecom, Baidu and other Chinese IT services. Scorpio is a modular, rack-server system with power and fans in six zones within the cabinet to make the rack easily configurable and manageable.

Based on statements from Intel and analysts familiar with its capabilities, the new datacenter strategy will look a lot like Scorpio, but will be tied together with a fabric interconnect that runs so fast—100gigabits per second—that the speed of the interconnect is more important than the location of the component.

It’s still unclear how closely Intel’s new strategy will adhere to the version Bryant has been laying out, but there has been little variation on the theme in any of Intel’s announcements or presentations to indicate any major last-minute changes.


Image: Intel