Fiber Mountain's Glass Core architecture is this month's winner of the Network Innovation Award. We caught up with Fiber Mountain founder and CEO M.H. Raza to discuss how its Layer 1 devices and orchestration software brings more flexibility to the physical network with SDN.
Editor's note: This interview has been edited for length and clarity.
Give us an introduction to the Glass Core architecture.
M.H. Raza: If you're designing a data center or a large enterprise network, you come up with a design for the physical infrastructure, you implement that, and then you put switches, routers, servers and applications on top of it. All the things above can be controlled via software, but the basic physical infrastructure has been extremely rigid. Once you put it in, for the most part, you just live with that forever.
Likewise, if you were building bridges in New York City, you'd decide where you're going to put the bridges across the various bodies of water. Once you put those bridges in, the rest of the city gets built around it and traffic is there. You live with your decision. If somebody built a football stadium on one side, traffic is going to be heavy on certain days. You can't change that.
Similarly, the network -- the physical infrastructure -- is that rigid. Once you put it in, then you live with it. And in the bridge example, if you wanted to reduce the congestion, you just are limited to controlling the flow of traffic. You're going to have signs and traffic lights, or tools in your mobile phone that tell you, "I think you ought to take a left and go to that bridge far away to cross over because this one's busy." You're just doing the control of packets -- or control of cars, in this case -- from a software perspective. You cannot change the fundamental infrastructure. If you had control via software to lay down six more bridges next to your bridge whenever you felt like it, how convenient would life be?
M.H. RazaCEO, Fiber Mountain
That's what the Glass Core does. For the first time, the physical layer is seen as a dynamic entity, changeable via software. We have a fiber optic network that can provide a connection from any port to any port in a matter of seconds, controlled by software.
SDN, in our opinion, is not complete without what we have to offer because a software-defined network includes not just Layer 2 and above, but also Layer 1 devices.
How does the technology work?
Raza: One of the fundamental pieces of equipment that enables this is the Optical Path Exchange (OPX), which is an optical cross-connect device. Now, optical switches have been around for a decade, but they've been expensive, power-hungry and large but limited in capability. They have been single-unit devices; it hasn't really been a network.
The OPX is a 160-port optical switch. Traffic coming in can be moved to any one of the other 160 ports. It's got very low power consumption and can do lots of multicast, but we've taken it further than that. A network of these switches can communicate with centralized software and determine how traffic will be routed across an entire network, rather than across a single device. These Layer 1 devices report to the AllPath Director, our central orchestration software, and can be configured via an SDN controller to deliver a connection.
To give you an example of how this works, let's say you have 100 servers and they connect to leaf switches [in a leaf-spine topology] or top-of-rack switches, which connect to larger switches and so on. The Glass Core can connect to any of the switches in the network or even directly to the servers. If a lot of your traffic from a server is going to another server cluster or storage, we can create a direct connection from a server port to a storage port and totally bypass the entire network. That reduces the load on the network, and there is very little latency -- five nanoseconds -- because you get that very direct connection.
What's been the alternative to this?
Raza: You needed human beings who had to be there in the data center. They needed to have a very accurate diagram of how things are connected. They would have to disconnect the cable, move it here and then pick it up over there to move it over there. But they don't want to run a cable across from left to right and up to down. They can't do that, because how do you document it? How do you know what you just added? You can't willy-nilly add something.
Our optical devices are in the middle and they're very dense, so all the cable connections are going through them. Via software, we can cross-connect anything that we want anywhere. And you can change it when you don't need it -- when you don't have that much traffic to that destination.
Does this create a single point of failure?
Raza: Absolutely not. We have the exact same redundancy level that any other network or architecture has. Everything has dual paths -- everything.
It's just that the paths are a lot cheaper. Let's say, for redundancy, you had to put two paths together and they were going through two switches. The switches have a lot of packet control, which means you have to manage the routes.
We have the same two paths, but the network is a lot simpler. There are no single points of failure, and we provide additional capability. For example, taps and monitoring -- our optical switches are capable of multicast, so we can take a stream of traffic that's coming in and see what's inside that stream. If it's malware, you want to analyze it. Non-intrusively, we can make a copy of that stream and deliver it at a moment's notice to any diagnostic device you have in your network. So we provide more security and monitoring inherently as part of the network.
With traditional vendors, if you need that level of monitoring, you build an entirely separate, parallel network to get the ability to do that. With us, there is no parallel network.
Orchestrating Layer 4-7 network services
What do SDN apps mean for Layer 4-7 services?
Why you should use physical and logical network designs