One way to look at Diamanti (formerly named Datawise.io) is that it builds an appliance that handles storage and networking for containers. But really, the San Jose, Calif., startup is all about the next step beyond UCS. Its goal is to rethink converged infrastructure for the big data era, where applications and their data can be spread across many nodes in a data center.
That hat trick of buzzwords — containers, converged infrastructure, and big data — makes it kind of fitting that the startup is announcing its $12.5 million funding round today, during the first round of the National Hockey League playoffs. CRV, DFJ Venture, GSR Ventures, and Goldman Sachs were the investors in the round, Diamanti’s Series A.
The 30-employee startup, which also includes personnel from Veritas and VMware, is announcing a beta customer as well — NBCUniversal.
Diamanti’s Non-DIY Box
“Build-it-yourself is fraught with peril,” says Jeff Chou, Diamanti’s CEO. “When you’re focused on topline revenue, you really don’t have time for your developers to be interfering with IT operations.”
What Diamanti really does is handle all I/O between containers and the network fabric. In other words the appliance is responsible for controlling containers’ communication with the outside world, thus keeping an eye on storage and networking performance.
Inside the appliance are CPUs and memory for running containers (no hypervisor involved), along with flash memory for short-term storage.
At first glance, you might wonder if the team could have done this with UCS. But UCS is designed to work with an ordinary SAN or NAS. Chou says neither UCS nor those storage networking options are built with more distributed applications in mind — the kind being applied to big data problems, in particular.
“We’re not built for [doing plain] virtualization, and we’re not built for Oracle SQL. That’s what UCS and SANs are for. We’re built for containers and microservices and scale-out applications,” he says.
Diamanti’s technology was developed in the form of I/O enhancements to Kubernetes, the open source container orchestration tool that originated at Google. Two of its contributions were announced in February and have been accepted for Kubernetes 1.2. One, named FlexVolume, is a means of automating I/O configuration; the other is a scheduler enhancement that takes storage and networking requirements into account.
In networking, Diamanti is concerned only with the data plane. The appliances are meant to use whatever control plane is in place.
That setup makes up for the lack of a coherent container networking model, Chou says. Even so, quite a few companies have begun tackling networking: Docker Inc. acquired Socketplane to flesh out the library now known as libnetwork; and startup Weaveworks is applying an SDN model to containers.
Photo: Nashville Predators wing J.P. Dumont (71) passing to Jason Arnott (19) for what’s about to be a hat trick. By Casey Fleser on Flickr, CC2.0 license. Photo has been cropped.