The Linkerd service mesh is doing away with its open-source delivery model — a significant change that Buoyant CEO William Morgan described as “the key to making Linkerd even better in the future.”

Rather than shipping new stable Linkerd releases in open source, the service mesh will be available only through Buoyant Enterprise for Linkerd (BEL), an enterprise-ready Linkerd distribution, beginning 90 days after this announcement.

While BEL will remain free for non-production use and for production use at organizations with less than 50 employees, larger-scale production use of Linkerd will require a monthly fee per cluster, paid annually. Early bird discounts and flexible payment terms are available, and Linkerd plans to accommodate nonprofit organizations and others with unique needs.

“It's always tough when someone says to you, hey this thing we've been doing for free for years now, well, now we need you to start paying for it,” Morgan told SDxCentral. “I'm very sympathetic to that situation, and we've tried our best to make this transition simple and low-stress for people, with a long grace period and exempting smaller companies and things like that,” he said.

The move away from open source is a necessity for Linkerd because it makes good business sense. “We’re asking companies that use Linkerd to invest back in it,” Morgan said.

This change is not driven, however, by sentiments like “this is only fair — you're building a big business on our hard work,” or “this is necessary for Linkerd to continue surviving in the first place” — however true they may be. Rather, this change is driven by the need for Linkerd to evolve.

“There are so many exciting areas we're going to build in Linkerd, and I truly believe we have an opportunity to 10x the value that people are getting from Linkerd today,” Morgan said. “But to do that, ultimately, we need the companies that are building their businesses on top of Linkerd to invest back in the project, and this is how we make that happen.”

Part of easing the transition away from open source is integrating a beneficial platform like BEL into the mix, which is particularly well-suited for sustained, large-scale production use.

As a service mesh creator that actually runs its own service mesh in production, Linkerd claims to have “a lot of painful visibility” into the needs of running a service mesh at scale. With BEL, Linkerd completed additional testing and hardening in certain environments and situations, and added tools that address concerns unique to sustained production use.

BEL’s new load balancer, for example, reduces cloud costs in multi-zone clusters by “being careful about where it sends traffic,” Morgan said. “That cost is something you only really care about when you're in a production situation where you're running lots of traffic through a highly available Kubernetes cluster. But if you're in that situation, you really, really care about it.”

Linkerd also introduced new mesh expansion capabilities in 2.15 that help limit the total cost of ownership and make Linkerd stand out from the competition. “As with all Linkerd features, it comes down to simplicity, specifically operational simplicity,” Morgan said.

The mesh expansion allows users to bring applications running on virtual machines (VMs), physical machines or other non-Kubernetes locations into the mesh. This ability to deploy Linkerd’s Rust microproxies anywhere outside of Kubernetes (K8s) (K8) and connect them to a Linkerd control plane running on a K8 cluster helps deliver a uniform layer of secure, reliable and observable connectivity between K8 and non-Kubernetes workloads.

Linkerd’s architecture is based around the microproxy, or an ultralight, mesh-only proxy rather than the large, general-purpose proxies used by other projects. “For mesh expansion, running those microproxies on VMs and other non-Kubernetes environments in order to mesh those new workloads is as simple as possible, and the team or platform engineer who is responsible for those proxies doesn't have to lose sleep worrying about whether they are going to suddenly run out of memory or go haywire,” Morgan said.