Let us cleanse our networking safe space by casting out some of the buzzword demons that make us dumber. Speaketh their name and let the cleansing fire of the divine light smite them!
- Software Defined Networking
- Workflow Orchestration
- Policy Driven Networking
- Hyperconverged Infrastructure Management
- “We have an API”
- Open everything
- Industry Standard Network Abstractions
And so it was that in a live presentation broadcast across the global Internet, Apstra shone their divine light into the Hall of Buzzword Bingo, and all marketing spin disintegrated in a fervent flash. More or less.
Even anti-marketing is a sort of marketing, but it’s marketing I can get behind. In this case, Apstra’s point is to stop focusing on labels, and start focusing on what it is you’re trying to accomplish with your network.
What Apstra helps you get done is deliver network state. That is to say, you as a network engineer design what the network is supposed to look like. Apstra handles the details for you.
Say you wanted to build a leaf-spine network. You’ve got some switches that can be used as building blocks in this design. Furthermore, the switches can run AOS (Apstra’s operating system) as a daemon. Wonderful. You can use the Apstra platform to build and configure that leaf-spine network for you.
The state you intend is for the network to deliver L3 packets via a leaf-spine design. You aren’t so concerned about the label on the front of the switch. You find no magic in configuring BGP paragraphs by hand. You don’t want to have to grok complex cabling schemes. It’s a simple thing that should just work.
Apstra tells you the cabling scheme. You connect the devices. Apstra programs the network devices with whatever is required to bring about your intended state. If the desired network state is no longer reflected by what’s actually running in production, Apstra will heal itself if possible, keeping you in the loop — your friend in day to day operations.
Wait just a minute…
Now, you might say, “Isn’t this just config management? Why are we getting all heady with the idea of network state? That sounds like another buzzword to be smitten by the divine light.” I, too, fell under this misconception, arguing with my friend Derick Winkworth and indirectly Apstra founder Mansour Karam about it. Emails were sent. Phone calls were made. But in the end, I think I got it.
Apstra is not about configuration management. Yes, configuration of network devices happens along the way in the paradigm most of us work in. But Apstra is after something much bigger. Managing the network as a single entity isn’t something that we humans can do. We are configuration managers. We roll out configs one device at a time. That’s how we think. As network engineers, too often our training was box-centric.
Apstra is about the network as a whole. Therefore, when you tell Apstra the network topology you’re looking for, the platform sorts out what that looks like, and makes it happen. Instructions are relayed communicating your intended network state. AOS implements the specifics implied by your intent. Call that “configuration” if you need to.
Even more interesting that bringing about network state is Apstra’s enforcement of that intended state. “Oh, well that’s what a network management station is for.” No. No, it isn’t. An NMS can alert you on a threshold exceeded. An NMS can raise a flag if the right regex parses a message flying by. An NMS might give you clues as to what’s changed. But ultimately, you, the human, are responsible to fix whatever broke. Not the NMS. Most NMS platforms are just dumb stat collectors with a reporting engine layered on top. Poll, throw the data into SQL rows, and barf up a graph or twenty. If you’re lucky, you’ll get a app dependency tree. This is still box-by-box management. There is no concept on an NMS of holistic network state.
Thinking about the entire network in terms of configuration stanzas and SNMP traps is all wrong. Yes, perhaps we’ve learned to stare at configs, assembling in our minds what the end result should look like. But that’s not a result that confirms that the network is, in fact, doing what we intend it to be doing. Figuring out whether it is, or is not, reflective of our intention is a job for software.
The view from the hot aisle.
Apstra is at version 1.1. The use cases are limited, and not yet ubiquitous. The 1.0 demos I’ve seen were aimed at designing, building and operating leaf-spine networks. 1.1 appears to have added host-routing and some other fabric attachment features. You aren’t able to build a WAN with it yet. You aren’t able to deploy a campus network yet.
Apstra supports a limited number of network operating systems. They cite Cumulus Linux and CVX, Arista EOS and vEOS, Cisco NX-OS, and Ubuntu servers on their products page.
Thus, Apstra is an entrant to a market that’s already seen competitors like Anuta Networks and Glue Networks make strong stands. Anuta and Glue aren’t doing the exact same thing as Apstra, but I find it exceedingly likely that an organization in the market for a product like Apstra would be evaluating Anuta and Glue as well. Apstra’s not going to be a no-brainer choice for everyone. They do a lot. The platform — the details of which I didn’t get into here — is modular, logically assembled, and ready to go with whatever sorts of use cases the market demands.
Obviously, Apstra is going for data center operators today. That’s where the volume play is. A few large customers with big fabrics to manage are wins for a product like Apstra. The network topologies tend to be more cookie-cutter and predictable as well, constraining the problem set for the v1.0 and 1.1 releases. That’s a wise play.
But what will the market demand, ultimately? Would I want an Apstra or something like it for my non-data center networks? My short answer is…you bet. And my reason is that it’s less about the building, and more about the operating. You don’t design and build networks all that often. But you do operate them constantly.
The idea of my friend, software, who understands what the network is supposed to be and watches it constantly for me is incredibly appealing. That need extends beyond my data centers and into my campuses, where the touch of human insanity is more likely to be felt.
I’ll be watching Apstra with great interest.