The data centre is heading into an era of “fabrics.” When most IT professionals hear about “fabrics” they either think of proprietary technologies, like QFabric, or perhaps of Fibre channel which is a storage technology. Cisco, Juniper, and Brocade have really taken to the term “fabric.” While there are certainly some new developments here – TRILL (transparent interconnection of lots of links) and DCB (data centre bridging) among them – what a fabric really means is a flatter Ethernet network, with greater link resiliency. A fabric network makes building a private cloud easier because it’s built for east-west traffic flows that are caused by virtual machines moving from physical server to physical server in search of more/less resources, depending on the automation and orchestration configuration of your private cloud.
So instead of thinking about “fabric,” really you can picture a flatter network – two tiers instead of three – that don’t rely on STP, where routing is technically being done a Layer 2, and routing can be based on location-independent L2 MAC addresses.
I recently attended a vendor-agnostic session called “Ethernet Fabrics 101” put on by Brocade in Toronto. It was really refreshing to see a vendor present basic facts about new technology with barely any vendor spin on it. Kudos to Brocade for that, and I’m looking forward to attending their next session.
Anyway, Brocade broke its definition for an Ethernet fabric into five characteristics:
- Flatter Architecture
- Distributed Intelligence
- Scalability
- Efficient
- Simplified Management
I’m going to spend the next few blog entries exploring those things, as well as TRILL, DCB, and where data centre networks are headed. 2012 is a pivotal year, but I suspect things will really take off in 2013 and 2014 as prices for 40GbE comes down, and even 100GbE for really demanding data centres (Apple, Microsoft, Google, Facebook, Twitter, and any cloud IaaS providers).