An Intriguing Approach to Modern Data Center Infrastructure

Source: https://gigaom.com/2020/08/27/nebulon-storage-composability-infrastructure/
August 27, 2020 at 09:06PM

I had the chance to meet with storage startup Nebulon recently and have been thinking about the company’s solution and value proposition. It is very difficult to figure out where they belong in the market, but in the end I decided that their product should be included in the composable storage category. Here’s why.

Nebulon’s product consists of two components: A hardware layer and a software system. Let’s explore these in turn.

The hardware layer is called the services processing unit (SPU), which is a disk controller on steroids that you install on a standard x86 server to provide sophisticated data protection and other services. This node can also be connected to other nodes and share resources to form a sort of virtual array called nPod. The controllers are PCI-based and connect through standard high-speed network interfaces. The storage resources in the back end, installed on the servers, are all-flash. This storage can be accessed directly from the servers themselves, like local disks. Servers can boot directly from the virtual devices or they can use these devices to store VMs or containers. This is all very cool.

The software consists of an intriguing SaaS-based management system. In practice, the cards are managed by a single SaaS application, a cloud-based controller plane called Nebulon ON, that can perform every type of operation—initial setup, configuration, you name it. It can even do firmware upgrades of back-end disks for all your cards, no matter where they are installed. The level of control and automation possible through this interface is very good, but it’s the API that takes it to the next level and enables users to operate the hardware at scale. A couple of demonstrations during Storage Field Day 20 really showed what is possible.

Composable Infrastructure: Why Nebulon?

I’ve already written about infrastructure composability, and I love the concept. Unfortunately, except for in a few niche use cases, it has been slow to take off. There are several reasons for this, including the fact that NVMe/TCP protocols are very new and NVMe-oF, in particular, needs expensive networking equipment to run properly. As I’ve said, the idea is good but we are not there yet, for practical and cost reasons.

On the other hand, a solution like Nebulon can bring very similar results, and more, at a fraction of the cost. The infrastructure can be simplified and standardized on ordinary x86 boxes, with cost-effective commodity Ethernet connectivity. All the magic is delivered via the cloud, giving the user a consistent experience with full control over large, globally distributed server estates.

Closing the Circle

Nebulon is not generally available yet. You’ll have to wait a few months to get your hands on it, and then it will be available only as a component in the configurator of your server manufacturer (and it is not 100% clear yet who those manufacturers will be). The idea is powerful, though, and I’m looking forward to seeing this technology in action.

At the same time, I don’t want to get too excited. Most aspects of Nebulon still need to be verified and further analyzed, but the idea has promise and I think it deserves attention from users looking for new technology to shrink costs and improve infrastructure operations.