Listen to the full podcast interview on Network Fabrics now!
Anyone who pays attention to data networking technology has likely heard the term “fabric” bandied about with respect to data center networks. To learn more about what fabrics are and the benefits they provide, we talked with Dhritiman Dasgupta, director of product marketing in the platform and systems division at Juniper Networks, a Carousel Industries partner and a leading company in this space.
Fabrics and the Evolution of Campus and Data Center Networks
For 25 or 30 years we’ve been building networks in campus and branch locations that are intended toconnect people to one another. Fifteen or 20 years ago, we started building networks in data centers with a very different purpose: to connect machines to one another, mainly servers and storage devices. But the technology we’ve used to build both types of networks was pretty similar: bridges, switches and routers.
But a lot has changed in the data center in the last 5 to 10 years with the advent of virtualization and cloud computing, Dasgupta says. At the same time, application architectures have evolved from mainframe-based to client-server and now to service-oriented applications. Put it all together and it’s hard for legacy network architectures to keep up with data center demands.
Wanted: A New Data Center Network Architecture
The problem is those networks were built mainly to handle “north-south” traffic flows that are typical of branch or campus networks, Dasgupta says. Traffic flows from one device, up to a data center or server, and down to another device. But data center traffic is more “east-west” – flowing directly from one server to another.
“The network is getting in the way,” when it comes to data center traffic flows, he says. What’s required in the data center is a flat, any-to-any architecture, where each device has a direct connection to any other. That’s what a fabric is intended to provide.
How to Build a Network Data Center Fabric
Companies like Juniper recognize that customers simply can’t afford to rip out all the existing data center network equipment they have and replace it with new fabric switches. So they are intent on ensuring new fabric technology works with switches customers already have, Dasgupta says.
The trick is in the control plane software. In Juniper’s case, its Q Fabric architecture can enables dozens or even hundreds of switches to appear as a single switch from a management and control perspective. That’s crucial to simplify management of all the devices, he notes.
At the same time, the strategy delivers the any-to-any connectivity that data centers need to be optimally effective. Essentially, the fabric flattens the network. Traffic no longer has to travel from one switch to one or more routers in order to get where it’s going; the fabric provides a direct, single-hop connection between any two devices.
Another essential element fabrics enable is location independence. With a fabric architecture, data center operations folks don’t have to worry when a virtual machine moves from one physical location to another. Because all devices have that any-to-any connectivity, it doesn’t matter where a VM is physically located in a data center; it can talk directly to any other server or storage system.
Dasgupta makes a good case that companies of all stripes will be migrating to fabric architectures over time, some sooner than others. Check out the podcast to hear our entire conversation with him and to learn more about fabrics.