New IP Agency

Competition Heats Up for SDN/NFV Processing

The shift to application and network virtualization by operators and service providers requires a new generation of multicore processors that are being introduced by many vendors. Cloud services are being delivered from hyper-scale data centers and through edge networks where support for virtual applications and virtualized networks requires servers with multicore processors that have many efficient cores, huge internal bandwidths and high-speed connectivity to memory and network interfaces. These requirements have driven the design of the latest multicore processors, which have been introduced during the second half of 2017 and into the start of 2018.

Server platforms supporting virtualized networking through SDN and NFV are replacing dedicated networking systems. These server platforms are installed in large data centers, located in telecom central offices and provisioned at customer locations, often as universal CPE (uCPE) systems. The virtualized software environments within all these server systems are similar and require similar processing capabilities, but at different levels of performance and integration depending on the location and physical environment.

The highest-performing server platforms are based on general-purpose multicore processors that integrate many high-performance cores together with large caches and high-speed memory and PCIe interfaces. These server platforms usually support dual socket implementations with two multicore processors. Network interfaces for these server platforms are usually provided through separate network interface cards (NICs). Network appliance platforms, uCPE systems and some other server platforms use integrated multicore processors which integrate the network interfaces and packet processing acceleration hardware, including security engines, onto a single device. These integrated multicore processors are also used for smart NICs and dedicated networking systems such as security appliances and gateways.

Heavy Reading's new report, Multicore Processors Accelerating Network Virtualization: A Competitive Analysis, identifies and analyzes vendors developing high-performance multicore processors for networking and server systems. The report profiles the leading vendors and analyses both general-purpose and integrated multicore processors, identifying their key features and highlighting the advantages they hold for service providers and telecom solution providers. The report includes not only information on the chipsets, but also offers insights into how the overall multicore processor market is developing.

The general-purpose multicore processors covered in the report integrate up to 48 high-performance cores together with up to 120MB of shared L3 cache and up to 8 DDR3 or DDR4 memory interfaces. Intel has dominated this market and recently introduced the Intel Xeon Scalable Processors (Skylake) that include microarchitecture enhancements for security processing and a new mesh architecture. Intel processors are being directly challenged by the new EPYC processors from AMD and Centriq processors from Qualcomm. IBM and Oracle also have general purpose processors being used in their own server platforms.

Most integrated multicore processors, like the Qualcomm Centriq processors, are based on ARM cores. This market has been led by Broadcom, Cavium and NXP, which have recently introduced new processor families. All three vendors have older multicore processors based on ARM, MIPS or Power architectures. Intel has also seen success in this market with the Intel Xeon D processors being used for smaller server applications and uCPE. Competition has increased significantly in this market with new high-performance processors from Marvell and Mellanox, and the Carlyle Group acquiring the X-Gene processors from Macom.

Most new operator and service provider investments are using SDN/NFV to deliver new services from data centers and through the network edge. The latest high-performance multicore processors have been designed with these applications in mind and take advantage of the additional memory and transistor count available with 10/14/16 nm technology. The new processor architectures have significantly increased internal bandwidths and processing capabilities and set up the leading vendors as they look forward to winning a significant market share.

— Simon Stanley, Analyst at Large, Heavy Reading

Simon_Stanley 2/23/2018 | 12:55:16 PM
Re: Project Denver becomes Ampere The ARM architcture is now widely supported and there is a good range of ARMv8 based processors with different integration and performance options that are very competitive on paper with Intel processors. The introduction of the Xeon Scaleable processors and Xeon D Skylake processors should however significantly increase the performance of NFV based solutions on these platforms. So far Intel has continued to dominate these platforms but there are a growing of companies with ARM based platforms and it will be interesting to see which platforms and processors their customers choose,
Gabriel Brown 2/23/2018 | 8:39:25 AM
Re: Project Denver becomes Ampere I read about Ampere and Renee James on Bloomberg (or similar type website). Interesting play.

Out of curiosity, how would you assess progress on ARM servers?

Fwiw, I'm hearing quite a bit about Skylake from vendors doing NFV performance tests and such like.
Simon_Stanley 2/23/2018 | 7:39:37 AM
Project Denver becomes Ampere The AppliedMicro processor business that was aquired by Macom with the rest of the company and then sold to the Carlyle Group has resurfaced as Ampere. Renee James, ex. Intel, is building a new business taking the AppliedMicro X-Gene 3 processor into production, as the eMAG processor, and building a large development team. For the company to succeed it needs to deliver production silicon that meets the original performance targets and show a compelling roadmap for the future.
Sign In