Hyperscale Networking Requires More than Just Bandwidth

Enterprise Networking Planet content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.

The data center is being pulled in two different directions as the enterprise strives for the perfect balance between hardware consumption and expanding data loads. On the one hand, on-premises infrastructure is getting smaller and more modular, while on the other, regional cloud facilities are pushing new boundaries in terms of size and scale and are becoming increasingly cheap and easy to use.

In the latter case, this is putting heavy pressure on data center networking technology, which must not only gravitate to near-carrier throughput and bandwidth, but must also juggle the competing demands of hundreds, if not thousands, of tenants. To accommodate both of these needs, the modern network needs to up its carrying capacity, as well as employ a raft of new routing and management capabilities.

According to BCC Research, the hyperscale market is looking at a 20.3 percent compound annual growth rate until at least 2022, when the market is expected to top $98 billion. As expected, much of this will be driven by leading cloud service providers like Amazon and Microsoft, but also by telecommunications firms, large financial service providers and others who require the same economies of scale and flexibility from their own infrastructure. Already, some facilities are supporting millions of virtual servers and related storage and networking gear.

This is expected to be the primary market for new wideband switches in the 100 and 400 GbE range. Companies like Broadcom, despite all the M&A turmoil of late, are pushing out new semiconductors like the BCM81724, which acts as a forward/reverse gearbox between 8×56 Gbps PAM-4 architectures and 16×25 Gbps NRZ environments. It also serves as a PAM-4 retimer to extend copper and optical links. The device features a low-power 16nm CMOS design and supports 40, 50, 100 and 200 GbE, as well as multiple 802 standards. The company is already shipping the device in volume, which means it won’t be long before it winds up in the channel.

Routing protocols are in need of a hyperscale upgrade as well, says Nokia’s Gunter Van De Velde. The IETF recently gave the industry some good news on this front with the establishment of the Link State Vector Routing (LSVR) workgroup aimed at crafting a scalable solution that meets the particular networking needs of the data center. These include support for equal-cost, multipath forwarding to replace complex ECMP, UCMP, IP Fast Reroute and other technologies. As well, it is expected to provide access to open source IGP routing without significant modifications to hardware switch stacks, and support broadcast-based routing at scale.

At the same time, the Open Networking Foundation has started a new project called Stratum that should make it easier for hyperscale data centers to build virtual networks on top of low-cost white box infrastructure. The idea is to develop a framework that makes it easier to program the data plane by extending control beyond just the data pipeline into configuration and operation of the network as well. This will be done by adding Google’s open-source gRPC network management interface and network operations interface for config and ops respectively. In this way, the group says it can support extremely lean operating systems and software that allow a wide variety of devices to tap into SDN deployments quickly and easily.

Players in the hyperscale arena are so large that they usually define and provision their own networking architectures rather than integrating solutions off the shelf. As these architectures trickle down to the mainstream enterprise, traditional vendors are caught in a bind because they have to transition from the relatively staid and profitable hardware market to the chaotic world of software.

But the writing is clearly on the wall. In order to maintain a healthy business for software-defined architectures in the enterprise, networking infrastructure will have to become easy and inexpensive to deploy and maintain. If this fails to happen, then the vast bulk of enterprise infrastructure will simply shift to the hyperscale cloud providers who have no need to pay top-dollar for someone else’s idea of what their networks should look like.

Arthur Cole is a freelance journalist with more than 25 years’ experience covering enterprise IT, telecommunications and other high-tech industries.

Get the Free Newsletter!

Subscribe to Daily Tech Insider for top news, trends, and analysis.

Latest Articles

Follow Us On Social Media

Explore More