One of the fundamental tenets to my research has always been that significant share gain opportunities happen during market transitions. New markets require new ways of thinking and different approaches, even in mature markets. This is how Arista Networks could enter the switching market and go from an idea to over $4 billion in market cap in just over 10 years.
One of the “new ideas” that Arista brought to the network industry was to build a highly scalable, robust cloud operating system that could be used as a foundation for rapid innovation now and into the future. The sweet spot for EOS has been large enterprises and big cloud providers with massive data centers, a market that has the most demanding performance and uptime requirements.
Despite the challenges of serving cloud providers, Arista has had tremendous success in the market by continuing to evolve EOS to meet the demands of cloud networking. Today, EOS is the only open, programmable state-driven, publish-subscribe-notify operating system designed specifically for the cloud.
This week, Arista announced the next phase of EOS. The upgrades are designed to enable customers to focus on migrating away from legacy environments to private, public, and hybrid cloud networks.
One of the big changes to EOS is the evolution from SysDB to NetDB. EOS is built on a Linux kernel, which allows each device to be managed as a server. SysDB is a centralized, in-memory database or file system. All the state that needs to be acted upon by EOS processes are stored in SysDB. As an example, the LED driver needs to be aware of the state of each port but not about the routing table, so it only loads the portion of the state it requires.
NetDB is similar to SysDB, but instead of providing system-level state information, it contains state information about the network to improve scalability, process efficiency, and resiliency. With NetDB, EOS supports over 1 million routes, 100,000 tunnels, and millisecond convergence, all keys for scaling the cloud. NetDB also gives the ability to aggregate the network state of all Arista devices to a common point in the company’s CloudVision platform. The centralized information can stream network-wide telemetry data to improve the efficiency of network operations visibility and historical analysis.
Another feature in EOS is what Arista calls tracers, which effectively trace virtual across the network by watching for a certain signature based on microbursts and changes in network performance. Year after year, Arista shows off VM Tracer at VM World, and there’s always high interest in it. Arista also allows customers to run virtual workloads as a network service.
EOS has extended its virtualization support by allowing customers to run Docker containers on EOS as well as trace the containers, similar to what Arista does with VM Tracer. The addition of containers to EOS will allow customers to run container-based applications directly on an Arista switch.
Arista has done a great job of leveraging software as a differentiator, and that positioning has allowed it to continue to differentiate itself on something other that just speed. The enhancements to EOS are a great example of the flexibility and extensibility of a software-driven approach.