We have spent some time explaining how Cumulus Networks is enabling the networking space to model itself after the Linux server revolution. The fruits of the revolution for the server space have been rather obvious, unleashing virtualization, clustered databases, and database orchestration while reducing prices and offering more options for hardware AND software innovation at a blistering pace.

So let’s examine the progression of the transformation in the server landscape and see if there were similar characteristics of inflection.

For servers, the incumbent mainframe style technologies often based the need for their products around:

  • Integer performance requirements
  • Floating point performance requirements
  • Transactions/second requirements
  • Graphics and disk performance

Up to the time that the Xeon line was released by Intel, people had dabbled with getting the base understanding of replacing mainframes with clusters, but were constrained by the factors above. This meant that the x86-based systems were confined to only certain small corners of a datacenter. Xeon turned this world on its head, by providing top of the line integer performance coupled with reasonable floating point performance.


Given the Xeon’s trajectory and economic promise, companies built high performance systems around it, making it a real contender against legacy solutions. At that point, the advantages of having popular and extensible operating systems like Linux, ESX, and Windows kicked the ball over the finish line. The clear indicator of the turnaround was that all major server players at the time built their latest product lines around the Xeon and slowly but surely deprecated their legacy operating system businesses.

So, why are we talking about this? Because the similarities extend to networking and for the Broadcom® StrataXGS® Trident II Switch SoC, which we believe is the product that will push the adoption of open systems over the tipping point.

Current incumbents using custom silicon often point to requirements like

  • Exotic buffer management schemes and protocols
  • Large forwarding tables with a myriad of attributes
  • High bandwidth and cumulative capacity
  • Support for a variety of protocols to implement features like redundancy
  • Support for low latency and so forth
Figure: The varied demands on a data center router

Figure: The varied demands on a data center router

The Trident II SoC, with 1280 Gbps FDX and starting at 500ns latency, addresses across the board all concerns about the performance of a single instance of switching silicon. Combine that with the elasticity we get with the Smart Tables and Smart Buffers feature of Trident II and you get a product that is versatile and can be configured to serve at any level inside a data center network. Mix in the port flexibility allowing any set of the 128 possible 10G ports to be combined into a 40G port, and you get a link that can be optimized for scale, capacity, or temporal congestion as per the need of the hour. Finally with VXLAN, we can divorce the complex, per community isolation and resource management needs of a data center network and manage that completely in the overlay context, freeing management and application layers to express their needs in an abstract way and letting the underlying network provide the service requested.

If indeed Trident II is the harbinger of the tipping point, and the indicators look good as all major OEM/ODM manufacturers are similarly lining up their premier products around its availability, then the next big factor to aim for is the operating system and the software ecosystem that it supports. The ease of use, familiarity, and robustness of this substrate will decide the pace of innovation for applications and solutions of tomorrow’s data center.

One casualty of such a change will be network controllers, which are needed to manage a disparate networking fabric with asymmetrical software interfaces. In this model, all boxes are made from an architectural building block. Thus scaling, much like for compute, is a function of horizontal replication of the building block.

The only questions that then remain are:

  • Which form factor will rule like the 2 socket x86 server did on the compute side?
  • Which OS will unlock the networking innovations and thinking like Linux vendors like RedHat, SuSE, and TurboLinux did for compute applications? Clearly, we think this will be us …