back

What Are Fabric Data Centers?

Greater demands from cloud-based applications require faster processing, which fabric data centers can help provide.

Fabric computing is a term that describes a tightly interconnected array of computer nodes, memory and storage devices that work as a cohesive unit. The grid-like arrangement of hardware provides a flexible, high-performance computing platform that can be reconfigured easily and scaled up or down as required. The interlocking nodes and closely connected parallel processing units make a diagram of the network look like a piece of cloth, hence the term “fabric computing”.

Standard (legacy) multi-tier network architecture introduces latency as clients and servers communicate along a path that resembles a tree. Latency increases when connections take place across several networks, or trees, especially when data needs to be collected from dozens or hundreds of different locations.

Faster memory and storage options are required to provide fabric computing networks with the speed necessary to work at full efficiency. Fabric data centers are designed to work with a flat, horizontal computing model that will eventually evolve into a single-tier architecture.

Eliminating multiple tiers is especially important when working with data-rich applications and cloud-based production environments that demand high bandwidth. Video meetings, IP-based telephony and on-demand entertainment have become commonplace, and as more cloud-based applications come to the marketplace, the need for faster processing and data throughput will only increase.

Data centers built upon the fabric computing model can add more processing power quickly and easily without creating an undue interruption to the current network. The simplified horizontal structure built into fabric data center architecture not only means easier implementation, it offers faster recovery from errors and hardware failures and requires fewer individuals to maintain the switches and servers.

Facebook Switches to Fabric Network

At Facebook, a significant amount of computing power takes place behind the scenes, outside of the user experience. Keeping track of posts and running non-stop games and applications creates a great deal of traffic. Targeted advertising and Facebook applications demand their share of bandwidth.

The on-going expense of adding additional hardware to the Facebook network was prohibitive, and doing so limited the IT department and Facebook engineers who had to depend on a small group of dedicated hardware vendors. Moving to a fabric-style data network meant changing network topology and overcoming a variety of challenges.

To implement the change to fabric computing, Facebook engineers decided to break up the network into identical, interconnected pieces they called “units of network”, or “pods”, each with 42 server racks and served by four switches designed for fabric network architecture.

Dedicated “edge pods” provide connectivity to the outside world. Facebook’s new fabric data network design balances load requirements and simplifies network construction and eliminates specific hardware requirements that limit the company to dealing with a small group of hardware manufacturers.

The modular nature of the new fabric data center design gives the Facebook network cheaper upward scalability and makes it easier to automate and operate than the previous cluster-centric network topology. The IT department at Facebook is free to work on other projects, with hardware upgrades taking up much less time than they did in the past.

Cloud-based computing and on-demand gaming and entertainment are motivating companies to continually re-examine and re-engineer the networks that serve businesses and consumers.

Information technology professionals will continue to be in high demand as fabric computing and network applications evolve. The market for support, software, and hardware necessary for fabric data center construction is expected to grow at a compound rate of 24% into 2019, with the majority of the growth taking place in North America, according to research firm MarketsandMarkets. That growth could propel the need for hardware engineers to design and crew these data centers, and work on the next generation of networks.

Explore Programs