Innovative 100GbE is the Solution for The Networking Bottleneck

For the last 30 years, Ethernet has been ahead of the curve in delivering the bandwidth and features needed for computing platforms.

It’s one of the key reasons that Ethernet has been broadly adopted across today’s datacenter and enterprise for decades.

But the network now faces the possibility of being a bottleneck for high-performance data center applications. In addition, new server and storage technologies are driving the need for higher network bandwidth than is commonly deployed today using 10GbE and 40GbE.

This change happened very rapidly. In fact, when 10GbE was ratified in 2002, it enjoyed a more than 6X performance advantage over SATA Rev. 1.0 -based storage interfaces and a 2.5X advantage over the fastest platform I/O standards such as 4Gb Fiber Channel.

Recently, storage bandwidth has increased with the use of new technologies such as Non-Volatile Memory Express (NVMe), and multicore Intel Xeon® CPUs have at least 16-lanes of PCI Express* (PCIe*) 3.0 connectivity, which at full capacity, can pump out more than 100 Gbps of data bandwidth.

Virtualization and cloud usage models are also adding to the throughput challenge as workloads increasingly spread across multiple CPUs and storage resources and must communicate frequently. This increases the “east-west” data flows between these resources, adding to the normal “north-south” traffic between the application and the user.

These trends have created a data deluge for network controllers and are driving the need for 100GbE in the data center. Intel is responding to this market evolution with the launch of the Intel® Ethernet Multi-host Controller FM10000 Family (formerly Red Rock Canyon), a next-generation 100GbE Ethernet multi-host controller technology that provides an innovative solution to this challenge.

The Multi-host Controller FM10000 combines several PCIe 3.0 interfaces along with 100Gb Ethernet ports and switch resources to provide flexible, high-bandwidth connections from multiple Xeon processors to the network. The FM10000 is designed to provide up to 200Gbps to Xeon processors at very low latencies, which will provide a dramatic performance increase for east-west data flows. In addition, it enables DPDK acceleration enhancements with its advanced frame processing pipeline and integrated tunneling engines.

If Ethernet becomes a bottleneck, it could impact overall system performance by preventing the full realization of compute performance as well as the full benefit of faster flash-based storage. As I wrote in a previous blog post, system performance is more dependent on the network interface than ever before – which makes the need for innovative 100GbE solutions critical.

Today, however, it will require both innovation and 100GbE performance to restore the performance balance. Intel is bringing both of those to the market with the FM10000.