Behind LinkedIn’s ‘Build, Don’t Buy’ Architecture Direction
By Edward Gately
The LinkedIn network supports more than 25 percent year-over-year traffic growth in support of more than 500 million users.
Many network operators would have turned to a vendor to drive their architectural decisions. LinkedIn, however, decided to go in a different direction. To support growth while adding business value, LinkedIn is disaggregating its network, moving from vendor-driven to white box to entirely new platforms, such as Open19.
LinkedIn is working with vendors and the open source community to create an entirely new way of building a data center network at scale.
During his Software-Defined Enterprise Conference & Expo (SDxE) session “Behind LinkedIn’s ‘Build, Don't Buy’ Architecture Direction,” Shawn Zandi, LinkedIn’s principal network architect, will discuss the history, technology and business drivers behind the decision to build rather than buy their network stack.
In a Q&A with SDxE, Zandi provides a sneak peek into the information he plans to share with attendees.
SDxE: What do you think is the biggest barrier organizations face with digital transformation — technology or culture? And why?
Shawn Zandi: I would say it’s mostly about the culture, and more specifically the culture of change. For us at LinkedIn, as a fast-moving company, change and transformation are parts of our DNA. Culture is something we take pride in. We embrace the change and we know that for those who constantly want to improve, change is necessary.
We also need to consider that digital transformation requires structure and solid execution. If the technology is not as capable or precise as expected, the outcome can be disastrous. Simplicity, structure and competency are the other factors to make this transformation successful.
SDxE: What challenges have you faced in your company with transformation? Where have you had success?
SZ: For us, the transformation part is mainly around the scale. In the last few years, our data centers and infrastructure have grown by 34 percent per year to support what is now half a billion members in more than 200 countries and territories. This demand dictates a transformation in how we build and operate our infrastructure and our product.
There are two main challenges at scale that usually stand out: scaling the technology (one’s product or network depending on the subject of discussion) and second, scaling the operations of infrastructure or product (which translates into people and tools). Hence decisions become more strategic than a purely technological choice.
SDxE: In your presentation, you’re going to discuss “Behind LinkedIn’s ‘Build, Don’t Buy’ Architecture Direction.” Without too many spoilers, can you give us an overview?
SZ: Operating at large scale requires a philosophical change in how we plan, deploy and manage our infrastructure. The three core principles that have guided our infrastructure design and our strategy are openness, independence and, most importantly, simplicity. To scale and own our infrastructure end to end, and control our destiny, we try to find the most minimalistic and simplistic approaches in infrastructure engineering, which are easy to understand and explain. And, being less complex, easier to scale. The outcome of this strategy had led us to projects like Altair, Falco and OpenFabric that have motivated us to build our own networking stack. This also helps us reduce our dependencies on third parties and bring competitive advantage to our business.
SDxE: What technologies/developments are you most looking forward to using in 2017 and beyond?
SZ: Empowering network and infrastructure to current trends such as big data and machine learning. We are building our own control plane for data center fabric, which I named OpenFabric. The idea is to have better traffic distribution, control and visibility inside our environment. OpenFabric is LinkedIn’s approach to a self-defined programmable data center that we will explain in detail in future publications. We are also working on an IETF draft to present this work as an internet standard so that everybody can take advantage of it.
SDxE: What do you hope attendees gain from your session and the overall conference?
SZ: The essential part of our discussion will be on scale, and what the plans or strategies are to make you ready to operate at scale. Speaking of the size and magnitude, there are many properties that should be considered when thinking about scale. I will bring a case study that we executed in LinkedIn as an example that might be applicable to other places.