Amid all the talk of software defined networking (SDN), Bandwidth as a Service (BaaS), and other big changes happening to enterprise infrastructure, it’s easy to forget that the little things still matter.
A prime example is the server and processor interconnect. As any data specialist knows, it doesn’t matter how fast a single bit traverses the miles separating the enterprise from the cloud: if the interconnect is overloaded, the application doesn’t get served or the data gets bogged down. In that regard, the old saw about the weakest link in the chain is highly appropriate as the enterprise becomes cloudier.
No wonder interest in the interconnect has revived of late. According to Research and Markets, the server interconnect market is on pace to see compound annual growth of over seven percent until 2016 as enterprises scramble to keep the advances in broader network architectures from hitting a brick wall in the server farm. And, as usual, where there’s a need, there’s someone waiting to fill it.
Mellanox to boost interconnect scalability
Mellanox is looking at new ways to boost interconnect scalability by placing hardware offload technology on the adapter. The ConnectX-3 Pro supports virtual overlay networks by offloading VXLAN and NVGRE processing from the CPU. This enables network architects to devise end-to-end scale-out infrastructure that is more in tune with the cloud- and service-based environments currently under development. Mellanox’s approach also flattens out the Layer 3 routing environment, which should simplify the migration of virtual machines across VLANs and SDN-based architectures.
Intel working on new silicon photonics tech to push interconnect speeds
The need for an improved interconnect doesn’t end at the server, though. Poor data exchange between processors can also hamper performance, especially considering the accelerating rate of processing capability. Fortunately for the enterprise, it’s likely that some intriguing work in the lab will bear fruit soon. Intel is working with new types of silicon photonics technology, work with the potential to bump interconnect speeds up to 100 Gbps. The company recently unveiled a fully functioning module consisting of modulators, detectors, waveguides, and circuitry, as well as a new type of hybrid silicon laser that the company says strikes a balance between speed, power, and cost.
Arteris continues ARM integration for better multicore interconnect
ARM processors are highly dependent upon interconnect speeds, since their strength lies in the ability to harness multiple chips for a single purpose. Arteris Inc. recently renewed its agreement with ARM Holdings for access to the new CortexA-7 and A-15 processor designs. The deal allows Arteris to integrate its FlexNoC network-on-chip technology deeper into the ARM architecture for improved multicore interconnect and other functions.
The interconnect may not be the most exciting piece of the emerging data puzzle, but it’s crucial for the smooth flow of information, not to mention the enterprise’s ability to meet the steadily increasing performance expectations coming from the data community. Wise network engineers will keep an eye on the micro architectures in the data center, even as most of the action appears to happen on the macro level.