With Intel’s Developer Forum (IDF) just around the corner, it’s a great time to assess the company’s current position and likely moves as it addresses tumultuous changes in its primary consumer and enterprise businesses. For IT managers, the focus should be less on PC and mobile processors, or whether Intel (finally) breaks into the smartphone and tablet markets, but the company’s strategy to become a more significant force, nay perhaps the dominant presence, within the data center. The IA platform in the form of its Xeon line, already rules the server market, with survey data from InformationWeek showing 95% penetration, with 8$% of IT organizations making extensive use of the platform. But Intel has steadily been moving up the value chain to thwart eroding commodification of its core business as white box servers have become almost disposable widgets for cloud providers while the cloud itself takes a larger share of enterprise business.
For over a year, Intel has been spreading its plans to re-architect the data center using a three-part strategy:
- Infrastructure optimization for greater power and workload efficiency
- Extending its processor and component designs to meet new workloads, specifically cloud, big data, HPC, network and storage systems
- enabling software abstraction layers in every part of the infrastructure and hastening the move to the automated software defined data center
Intel has been pushing x86 processors, which now span everything from high-end Xeons for mission critical systems to low-power Atoms repurposed from tablets to microservers, into more and more markets. This means the company’s engineers have been busy analyzing the gamut of IT workloads, which they characterize by two key attributes: CPU and memory usage versus I/O intensity. Common enterprise applications typically have very different CPU, memory and I/O requirements than cloud scale applications, big data/Hadoop distributed databases, multi-tenant virtualized server farms, custom HPC applications, storage systems, even network switching, packet control and network services collectively known as SDN and NFV (network function virtualization). But Intel wants to ensure there are chips, reference platforms and optimized software for all of them.
Intel’s expansive strategy is leading the company to augment the base IA platform with special purpose communication processors, network accelerators (for crypto and compression) and parallel processing coprocessors (Xeon Phi). Ever since the Fulcrum acquisition, Intel has also redoubled efforts to become a significant force in the merchant switch silicon market long dominated by Broadcom and others. At IDF, look for Intel to continue forking and supplementing its core IA CPUs into derivative products, whether different SKUs of discrete processors, customized SoCs, motherboards with multiple components or full-fledged reference systems, targeting the panoply of data center infrastructure. Intel likes to tout its technological lead, architectural consistency and software compatibility, however whether these, along with much improved performance and workload-optimized platform, will be enough to win the business of big systems houses like Cisco and EMC or Dell and HP (outside their server businesses) is the open question, but given Intel’s obvious technology and semiconductor manufacturing prowess, it has a decent chance.
The ‘Intel Inside’ Data Center
Intel wants to extend its dominance in the Windows and Linux markets into the era of cloud computing, XaaS, converged data and storage networks and 10 gigabit (and beyond) networks. With PC sales in a seemingly permanent funk, It’s not enough for them to just be in your laptop. The goal is clearly Intel Inside everything in your data center. The company is overdue to bring its Haswell microarchitecture to the server line, indeed Dell and Lenovo have already spilled the beans with new workstation specs that include the v3 chip. Although we won’t know architectural and performance details for a couple weeks, expect to see the usual improvements in instruction execution and power efficiency, memory performance (the company has already confirmed next-generation DDR4 support for its enthusiast consumer line, and with faster, more power-efficient DDR4 memory already shipping, it’s natural to assume it will debut on the new Xeons) and network throughput, bolstered by its new Fortville switch silicon (again, the company has confirmed its next generation 10/40GbE controller, which will include network virtualization offloads for NVGRE and VxLAN, will ship with the new Xeon platform).
Yet as Intel’s enterprise business goals have gotten grander, don’t expect a typical ‘speeds and feeds’ processor refresh. In what amounts to the semiconductor version of mass customization, IDF announcements will almost certainly include several variants of the base Xeon, some paired with specialized co-processors, accelerators and/or reference boards each targeting a different data center workload to include:
- traditional enterprises
- hyper scale cloud service providers
- HPC and highly parallelized workloads
- conventional SAN storage arrays
- converged/scale out distributed storage
- network switches.
And those will be just the standard parts and reference designs available to anyone. Last year, the company disclosed joint development projects with mega cloud services like Facebook and eBay to develop custom SoCs based on standard Intel designs. This allows tech-savvy big buyers like the major cloud services to augment standard x86 cores with custom accelerators, power management hardware or other features to optimize performance for specific workloads. In essence, Intel has turned the Haswell core into a reusable macrocell that can be repurposed into systems designed for very different applications.
Still an open question is Intel’s commitment to low-power, high-density systems using its Atom processors, which are now facing serious competition from the first generation of 64-bit ARM CPUs. Although Intel continues to develop the Atom core and shows a next-generation Denverton platform on its roadmap, the continued emphasis on Xeon and ‘big core’ Haswell and Broadwell architectures, along with the upcoming Broadwell SoC ‘twiner’ product initially targeted at notebooks, but easily repurposed to servers, indicate the company would prefer customers continue to gravitate towards its more expensive and higher margin product lines.
Intel’s standard response to the big-versus-little core debate is that virtualization allows big core devices to scale down to little-core workloads, while also providing the versatility to accommodate more intensive workloads. This message will continue to resonate with enterprise customers, however cloud services and hosting firms may drift towards microserver platforms now that 64-bit ARM provides ‘good enough’ performance, higher memory capacity and virtualization support. Thus, we may see the market further bifurcate between traditional enterprise IT and cloud-scale services buying servers and switches by the pallet load.
I’ll have more to say about Intel’s plans and prospects after IDF, but next week it’s off to VMworld in San Francisco, the IT industry’s answer to Burning Man where tech analysts and IT geeks gather to see what’s next for the software abstracted and automated data center. Watch for updates and look me up if you are at the show.