Everywhere you look, there are technologies and products intending to transform networking. That’s because networking has remained a box-by-box architecture for more than 25 years – stable and proven, but complex, proprietary and expensive. But modern forces of open hardware and disaggregated software together are driving an irreversible transformation in the datacenter today. To make networking faster, simpler and more economical, one should look no further than the advances hyperscale organizations such as Google and Facebook have made in their own environments with open networking hardware and software innovations.
At Big Switch, we were inspired by the hyperscale players, but have created an enterprise SDN fabric that brings networking into the 21st century to become a first-class citizen in the data center, alongside compute and storage. We are able to create hyperscale-style network architectures for mainstream enterprise and service provider customers by leveraging open switch hardware and SDN-based leaf/spine fabrics.
We have already witnessed a transformation in compute where the mainframe – a massive and proprietary computer – was disaggregated into an x86 based supercomputer (see Figure 1). The key to this disaggregation was standard, open server hardware and an intelligent virtualization software layer to optimize compute utilization and operational agility. We achieved tremendous innovation velocity and dramatic cost reduction, as compute innovations occurred independently across HW and SW.
Figure 1: Compute Innovations with Disaggregation of Mainframe
Can a similar approach also be applied to networking to disaggregate the current defacto industry standard network mainframe chassis architecture: Netframes. Netframes are massive and proprietary network chassis – basically the mainframe of networking (see Figure 2). To disaggregate the Netframe, switch hardware needs to become open, so that third party software can run on it. The Open Compute Project (OCP), led by Facebook, is driving network hardware openness and Software components, such as Open Network Installation Environment (ONIE) and Open Network Linux. This enables white-box and branded white-box (brite-box) switch suppliers to rapidly enable third party network OS and SDN stack to drive innovation velocity independently across networking HW and SW.
Figure 2: Opportunity to disaggregate Netframe chassis
The key point to suggest Netframe disaggregation lies in hyperscale network architectures where networks have been faster, simpler, better and economical. An example is Google’s Project Jupiter which has leveraged merchant silicon based open hardware, software based controls and a leaf/spine switch fabric to build data center networks (see Figure 3).
Figure 3: Google’s Project Jupiter principles for DC Networking
These three Jupiter principles can be applied to disaggregate the Netframe. As any networking professional knows, a switch chassis has three canonical components: a supervisor for centralized control and management plane on the supervisor, fabric backplane and Ethernet line cards (see Figure 4).
Figure 4: Netframe Disaggregation into an SDN fabric (Logical Redundant Chassis)
So to disaggregate the Netframe:
Extract the supervisor function and place it onto an x86 server – this becomes the SDN controller.
Fabric back plane can be extracted and placed as the spine layer.
And finally chassis line cards become the leaf layer, or top of rack switches.
Effectively the Clos architecture within the chassis controller by supervisor is now disaggregated with leaf/spine clos fabric with standard 1RU 10G/40G switches controlled by an SDN controller residing on an x86 server. This SDN fabric is essentially a redundant logical chassis – a redundant and disaggregated Netframe – but now provides scale-out deployment.
Unlike the chassis supervisor, an SDN controller coupled with open networking HW can take advantage of Moore’s law by riding the x86 server hardware evolution. Additionally, with merchant silicon based open networking hardware, whether it is whitebox or britebox, Netframe disaggregation leads to HW vendor choice and lower CapEx.
Big Switch offers two types of SDN fabrics for data centers: Big Monitoring Fabric (BMF)* and Big Cloud Fabric. Big Monitoring Fabric is a next generation network packet broker for tap and span aggregation to enable pervasive monitoring and pervasive security. Big Cloud Fabric offers a pod fabric for OpenStack and VMware workloads for public/private cloud, VDI, analytics, software defined storage, and emerging container environments.
These SDN fabrics are out-of-the-box and ready-to-deploy solutions, there is no do-it-yourself experimentation required. And they are being deployed in production at scale by mainstream enterprises and service providers.
Market reality on transforming legacy box-by-box architectures is very clear: (1) open networking train has left the station, and (2) chassis Netframe disaggregation via open SDN fabrics is underway. Customers like it because their networks are faster, simpler, better and economical.
Welcome to the 21st Century of Networking!
Vice President of Product Management & Strategy
* Formerly Big Tap Monitoring Fabric