Microsoft Goes Barefoot

Chip2

By: Michael Vizard

One of the least appreciated aspects of cloud computing is how major cloud service providers such as Microsoft (MSFT) are dependent on a network operating system (NOS) to specially address the requirements of a hyperscale IT environment.

In the case of Microsoft, for example, the Azure cloud makes extensive use of SONiC, an open source NOS based on a modern microservices-based architecture that Microsoft built using Docker containers. In contrast to how most enterprise IT organizations consume networking technologies today, that NOS is designed to run on disaggregated network hardware that serves two purposes. The first is that it allows the internal IT shop running the Azure cloud to more flexibly respond to changing networking requirements using white-box hardware. The second is that it also serves to keep Microsoft from getting locked into any specific hardware vendor.

At the Open Compute Project (OCP) Summit this week, Microsoft in collaboration with Barefoot Networks, a provider of a switch based on a custom processor that can be programmed using the P4 language, demonstrated the capability to support bare-metal server hosting along with providing access to sophisticated telemetry data. The increased telemetry capabilities include monitoring of network performance scenarios, such as path and latency tracking, microburst detection, and congestion analysis. 

SONiC for Microservices

Microsoft makes SONiC available as an open-source project that other organizations can use. But even if IT organizations don’t want to support SONiC on their own, they might pay close attention to how Microsoft and cloud service providers are transforming networking, says Arkadiy Shapiro, product line manager at Barefoot. Many of the concepts being pioneered by cloud service providers, such as microservices architectures, will soon manifest themselves in NOSes from traditional networking vendors, says Shapiro.

That transition, notes Shapiro, will be critical in terms of enabling network operation teams to more adroitly respond to not only provisioning networking services, but also giving applications more control over the networking services they consume. In fact, Shapiro notes that NOSes based on microservices have a very symbiotic relationship with programming languages such as P4.

“P4 and microservices are a perfect fit,” says Shapiro.

Programmable Networking Trend

That programmability will also play a critical role in the melding a traditional NetOps and emerging best practices for DevOps that are just now gaining momentum within enterprise IT organizations.

It’s not clear to what degree IT organizations will eventually decide to separate the purchase of NOSes and the underlying hardware that supports them. But there’s been enough interest to require traditional enterprise IT networking hardware vendors to provide support for disaggregated networking gear as an option for customers.

Regardless of whether IT organizations continue to buy software and hardware from the same networking vendor, the one thing that is clear is networking will soon be catching up with rest of IT in terms of agility -- and programmability of the network is key. Most IT organizations today can provision a virtual machine in a matter of minutes. Providing networking services on-premises in most cases still takes days, sometimes weeks. That inability to respond to requests for networking services is one of the primary reasons so many application developers prefer to rely on external cloud services. The provider of those services built their own programmable networks based on software-defined network (SDN) running across disaggregated network hardware that enables them to provision networking services in minutes.

At this point, it’s now more a matter of when, rather than whether, these same programmable capabilities will come to local data centers. Much less clear, however, is how they arrive and via what vendor.