Open-Source Networking? How Disaggregation Found Its Way Into the Data Center

Here you are. You have all documentation ready and in-hand to present to your manager. The year is 2004, and you are pitching a new idea for updating the company’s existing network core infrastructure. The terms “cost-savings,” “scalable,” and “extensible” are prevalent within the proposal documents. You are unsure if the terms “open-source” and “networking” were ever mentioned in the same sentence before. However, the solution is simple: build a network that is non-blocking, resilient, predictable, and manageable.

Additionally, you adopt and include the operational efficiencies seen within your SysAdmin department. Who doesn’t love streamlined provisioning, broad visibility into resource utilization, and scalable deployment of services? What if we could provide the same network technologies across a variety of hardware platforms?

It turns out we can, using open-source networking to decouple hardware vendors from software. While this is only one option available in the market and may not work for every enterprise organization, the disaggregation of network hardware and software has some definite advantages.

What is Disaggregation in the Data Center?

To compare the above situation with today is a little unfair. The concept of marrying network and system operations may have been left to universities for research and rarely seen in the enterprise network back then. Continual advancement in merchant silicon has made it possible for many companies to step away from the clutches of closed-source vendor solutions. Solutions have sprung up to disaggregate the network operating system (NOS) from specific vendor hardware. Imagine using the same software, — but getting to shop for the network hardware on which to install it!

What is disaggregation? We must first define “merchant silicon” and its place in the data center. “Merchant” or “off-the-shelf” chip sets are available to the world . Bare metal networking switches  allow software companies to program the latest and greatest in standard network technologies (Overlay, Layer 2, Layer 3 protocols, MC-LAG, etc). Previously — and with many network equipment vendors still operating today — you would have to use the software provided by the vendor, which could lead to limited network design potential with your specific topology.

With disaggregated hardware and software, engineers can design a highly-available, non-blocking, and efficient network architecture without the cost implications of a closed-source solution. Such programmability and availability provides a number of benefits:

• Competitive pricing compared to “in-house” network vendors
• End-users can use existing open-source applications/tools to interface with networking devices
• A company provides support for its own NOS, while outside groups and communities also help fix bugs, write new code, and generally provide a great resource for questions. In many cases, developers contribute on both sides! This is one of the key advantages of open source philosophy in general.

No longer do we need to work at Facebook/Google/Amazon to experience the flexible nature of open networking. Disaggregation extends our choices from the systems side into the networking world. We can invest in a NOS solution but choose what hardware to use. Blending the approach of systems and networking allows for greater operational return.

Sounds Great! What’s the Catch?

As with anything in IT, there is no one-size-fits-all, and the concept of open networking may not make sense in all scenarios. This is a relatively new approach and typically seen in a data center, where the focus is primarily on efficiencies in East-West traffic flows (traffic that flows between hosts within the datacenter and stays there, compared to North-South traffic, where flows go in and out of the datacenter.)

More importantly, closed-source solutions have their place and will for a long time. Proprietary hardware and software vendors provide the benefit of thoroughly tested solutions that have worked well for many years. Most customers will not do a complete rip and replace in favor of solutions that have not been in the market for very long. A balance must be made, and any approach should be carefully driven by business goals.

We are, however, in the midst of a very exciting time in networking. Extending the long-standing and community-based creativity of the open-source world into the network is the bullet train this generation needs to survive.

Posted By: Cloud Services Engineer I Alex Kirby