Last week I attended Networking Field Day 7, and was introduced to Pluribus Networks. Pluribus is taking an interesting approach to building the data center fabric, by combining high-performance data center top-of-rack (ToR) switching with powerful server internals in a platform they’ve dubbed the Freedom Server-Switch.
Server Brawn + Switch Brains = Infrastructure Fabric
The Freedom platform can be loaded to bare with RAM and storage along with some pretty powerful CPUs (this data sheet provides all the details), which enables embedding various network (and not-so-network) services right in the network at every edge. The platform runs the NetVisor operating system, based on BSD. This software can be had in various feature levels:
Various services that can be enabled beyond typical L2/L3 network services include DHCP, DNS, PXE, load balancing, CDN functions, NAT, NAS (yes, really), and traffic analytics. Since these switches are designed for deployment as leaf nodes in leaf-spine architecture datacenters, this embeds these services right at the network ingress point for each connected device.
You may be thinking about the potential administrative overhead included with performing advanced network services on each ToR switch, but that burden is eased with fabric-wide management features that allow an administrator to interact with any node in the fabric and issue commands that can affect a subset of fabric nodes, or the entire fabric at once.
During the NFD7 demonstration, Pluribus Networks CTO Sunay Tripathi showed us the ease with which the entire fabric (the Fabric Cluster, as they called it) could be programmed to single out a specific traffic flow (based on any number of parameters), and perform some operation on it such as redirecting it to a specific port, a service running on the Freedom platform, or copy the traffic to local storage. With a couple commands, he was able to intercept and store traffic matching the flow parameters from anywhere on the network the flow may appear. This was powerful stuff. And of course, since Pluribus exposes APIs for accessing these features, one can imagine the ability to automate various network service functions from external applications. In fact, Pluribus provides an SDK for “bare metal” access to the switch so that future applications could potentially extend functions beyond anything that’s been thought up so far. Additionally, VMs can actually run on the platform, so perhaps other functions traditionally centralized in the network (IDS/IPS, anyone?) can be embedded right at the network edge.
Something that really struck me about the Pluribus NetVisor software was that the fabric was equally manageable from a Unix command line, a rich switch CLI (although the syntax looked quite a bit different from anything I’ve ever used, so there’d be some learning curve there), a web-based GUI called vManage, and a variety of API interfaces. Lately, the industry has been laser-focused on APIs, APIs, APIs. I thought Pluribus struck a good balance with their approach recognizing that the CLI is not dead, and APIs provide another, but not exclusive, vector for network management. These various tools could be leveraged by network administrators that are comfortable and adept with different administration models and none appears to be handicapped by their choice.
More than that, though, what I saw in Pluribus’ platform was a bold attempt to move toward what may well be an inevitable future. I’ve been thinking for some time about how in the not-so-distant future as network, compute, and storage facilities coagulate we’ll not have many “network engineers” or “server engineers”, but rather “infrastructure engineers” who know how to work everything. Sure, we may still retain a focus or specialty, but it’s going to become very difficult to claim “I’m a network engineer. I just provide the network. Servers and storage aren’t my thing.” At least, if you want to stay relevant and have a job, it will be difficult.
The Pluribus Freedom Server-Switch really embodied that notion. Rather than building a high-speed switching fabric that has services blocks hanging off of it to provide network services, application services, storage, security, monitoring, and even applications themselves, the Pluribus solution struck me as an infrastructure fabric, providing many of those services right in the fabric, at every point of ingress and egress. Surely Pluribus is not trying to replace enterprise or tenant servers themselves, but moving the various utility services into that infrastructure fabric consolidates the deployment, administration, and management of those infrastructure support services allowing the servers and storage attached to the fabric to be used for what they’re intended for — applications.
While I saw a lot of promise in the Pluribus Networks offering, I do think they will have a bit of an uphill battle in many shops that have not yet moved to a more consolidated “infrastructure team” approach (which is most environments I see), as the server and storage teams may feel threatened by the idea of “the network” running various services and even hosting storage. I suspect this technology will be a better fit in more agile environments that have embraced a holistic approach to infrastructure services.
I strongly recommend watching these videos from Networking Field Day 7 as they really demonstrate the fascinating approach Pluribus Networks has brought to the table. Pluribus Networks also has some good whitepapers sprinkled around their site that are worth a read as they present some good technical detail rather than just marketing fluff.
Disclaimer: Pluribus Networks was a sponsor of Networking Field Day 7. At no time did they ask for, nor where they promised any kind of consideration in the writing of this review. The opinions and analysis provided within are my own and any errors or omissions are mine and mine alone.