Exploring How Disaggregation may change the future of R&E networking.
Having developed and operated nationwide WAN services built upon OpenFlow over the last 5 years, the GlobalNOC experienced the challenges introduced by an imperfect protocol incompletely supported by traditional networking hardware vendors. With NSF support, we began to look more closely at the emerging trend to disaggregate hardware and software known as Whitebox switching.
This site shares the results of our explorations and the implications for the R&E networking community.
What is Whitebox Networking?
The term originally referred to Switches build with Merchant Silicon, switches built using Switch ASICs not developed by the switch vendor. Often this type of hardware was custom built by Original Device Manufacturers(ODM) for large scale buyers who were Switch Vendors.
Over time large buyers of these equipment began to engage the ODMs directly and projects to define open hardware standards to facilitate the commoditization, disaggregation and collaboration were created.
Today we see much effort in the data center but also movement to support WAN use cases with most projects using linux as the base environment and providing various APIs to control hardware features.
Why Disaggregation?
Network hardware today is typically vertically integrated, you buy a Cisco switch you run IOS. A potential challenge to innovators stems from the hardware vendor dictating management, control and data plane options. While there are many advantages to the integrated approach, there have been several industry efforts to develop alternative approaches which support innovation both in terms of feature set and vendor economics.
The goal of disaggregation in general is to decompose the essential hardware and software components into logical building blocks for recombining as need to provide a more ideal balance of cost and capability for a particular deployment scenario.
What is our Role?
This project is funded by the NSF through EAGER grant #1535522. We work with R&E community stakeholders and ecosystem participants to identify and evaluate use cases and network components in our lab, sharing the results of these efforts with both.
Feel free to contact us with any question or use-case that you think might benefit from this approach to networking. We are very interested in understanding your use case, guide you though the options bare metal networking provides, and set up a PoC in one of our testbeds.
You can reach us at:
What benefits can the R&E community expect?
There are several areas we feel will be of value to the R&E Community that stem from an open architecture based on merchant switch silicon.
- improved vendor economics through a set of solutions tailored to community use cases that may converge components from layer1 through layer3 as needed
- scalable Linux based management techniques with ability to converge some forms of NMS function directly onto switch. For example running active network probes.
- new measurement capabilities based on emerging switch silicon that allows for the direct observation of buffer use and prevalence of microbursts, a key concern when selecting network hardware destined to support large data transfers.
Additional Topics
And why was it developed?
A look at goals and architecture of the Microsoft SONiC switching stack used in the Azure Cloud
The different approaches and the consequences
What makes a fabric so different from traditional approaches?
Why is it so important for enabling whitebox / bare metal networking?
How to get to flexible hardware pipelines.
Integrating DWDM and the routing stack
Linux integrating the networking hardware.
