Long-rumored to be the "OpenFlow company of the Northeast," Massachusetts-based Plexxi is coming somewhat out of stealth mode with a product set that has little to do with OpenFlow, but takes on data center fabrics with a new kind of software defined networking (SDN) strategy.
By submitting your email address, you agree to receive emails regarding relevant topic offers from TechTarget and its partners. You can withdraw your consent at any time. Contact TechTarget at 275 Grove Street, Newton, MA.
Plexxi's technology makes the most of the pipes available in the data center so that the components of an application can talk to each other when they need to. While perhaps that sounds similar to the problem Juniper with QFabric, Cisco with FabricPath, and Brocade with VCS are tackling, Plexxi is doing something those guys aren't, and probably won't: dynamically figuring out the end points that must talk to each other, and then ensuring that the network path (with your choice of network path characteristics) is available.
The problem with network fabrics
Read more Fast Packet blogs
Cisco wires it's own concept of software defined networking
HP OpenFlow: A move away from the 'tyranny of CLI'
Network fabrics purport to enable any fabric-attached endpoint to talk to any other fabric-attached endpoint, without bandwidth or topological constraints. But even in a fabric topology, there are chokepoints in the network. There's contention where different conversations are fighting to be forwarded at the same time. Storage fights with applications. Conversational bursts fight with other traffic already on the wire. Network engineers know these problems well, and the traditional solutions have been to either build the pipes bigger, use protocols to manage contention, build isolated network segments that can be dedicated to specific applications, or cleverly uplink hosts in order to minimize traffic flows outside of the network pod. But once you throw virtualization into the mix and start moving virtual machines around, the problem gets much harder to resolve unless you have an infinite amount of money to throw at bandwidth.
How Plexxi handles the problem:
- An Ethernet switch: You might be thinking, "Oh, boy, another Ethernet switch, in a market that's already saturated." And you're not wrong. There are so many switches on the market that some of the larger vendors are competing with themselves for a share of coveted data-center rack units. But Plexxi's switches are plumbed together using a unique optical interconnect. It's across this interconnect that Plexxi manipulates traffic flows going between switches. How? In optical networking, it's possible to create multiple virtual circuits across the same physical piece of fiber optic cable using different wavelengths of light called lambdas. So, instead of scaling interswitch bandwidth by adding physical interfaces and cabling that are plumbed in a mesh, star or leaf-spine topology, a Plexxi network scales using lambdas over an optical interconnect.
- A controller and API: Plexxi switches have an API, and a Plexxi network includes a controller that writes to the API. Plexxi' s controller gives you the ability to manipulate a Plexxi network into prioritizing traffic in particular ways, and there is an application that can do that for you. Customers that wish to forward traffic according to their own paradigm can write their own application that talks to the controller.
- Affinity groups: Plexxi has published a white paper describing the notion of an affinity group in the context of networking. A networking affinity group is just what the name implies -- a group of network-attached devices (both physical and virtual) that need to talk to one another for whatever reason. They have an affinity. This makes sense, as most applications have several components, typically residing on different devices. For example, an application that serves content to a Web browser probably has a database, middleware, HTTP engine, load balancer and perhaps a proxy server. In an ideal world, the network would be optimized such that all devices the application depends on for its performance can talk to each other in a prioritized manner. Plexxi allows for the creation of an affinity group that is dynamic, such that the network forwarding and prioritization can be tweaked so application delivery has a network guarantee.
Plexxi's solution then, while like the OpenFlow controller model in some respects, is not limited to the OpenFlow specification or mere programming of flow tables. Instead, Plexxi is going for a network paradigm that will allow for software-based, controller-driven, API-centric customization of a data center network to make sure the network isn't getting in the way of applications doing what they need to do.
There's more to come from Plexxi, and all will be revealed as the company emerges fully out of stealth mode. It's sure to be interesting. We'll keep you updated.