CIOs are faced with a never-ending barrage of new technologies; understanding which are relevant is a big challenge. As I wrote recently, networking solutions are moving towards “stacks and fabrics” to support virtualization requirements. Growth and scalability are top considerations for choosing a networking solution. While almost all data centers need to consider the impact of virtualization on network design, most users are well served by precursors to fabric architectures, namely switch clustering, today.
Cisco recently announced an update to its networking architecture including what it claims to be the most scalable data center fabric on the planet. The reality is that the average data center only has 500 servers and while growth is a big challenge, server virtualization has slowed down the sprawl of physical servers for many customers. With virtual environments, customers need networking that can provide high bandwidth between all ports on the network. Long before the current wave of fabric technologies, networking vendors were offering ways to supersede Spanning Tree Protocol (STP) through switch clustering technologies that allow two or more switches to be managed and act as a single switch.
A Cisco blog post from the recent announcement lists these solutions as “box scale” offerings. Cisco has been offering these types of technologies (specifically vPC and VSS) on both the older Catalyst and newer Nexus product lines for years. Below is a table of current switch clustering and fabric technologies and the number of current installations.
| Networking Vendor | Switch Clustering Technology | Fabric Technology (vendor data on installation numbers) |
| Arista Networks | MLAG | TRILL (future) |
| Brocade | vLAG | Virtual Clustering Switch (hundreds of customers since introduction in ‘10) |
| Cisco | vPC/VSS | FabricPath (over 500 customers since launch in ’10) |
| Dell Force10 | VirtualScale | TRILL (future) |
| HP | IRF | TRILL (future) |
| Juniper Networks | Virtual Chassis | QFabric (recent GA, less than 10 installed) |
Most of the networking vendors consider the two types of technologies to be complimentary. HP recently announced an update to its IRF solution, expanding its support to 4 12500 switches (formerly the A12500 that supports 128 ports 10GbE or up to 512 ports with 4:1 oversubscription). HP has stated that it will support mixing IRF and TRILL in a future release. While Juniper only recently started shipping its full QFabric, Virtual Chassis has been deployed in thousands of environments and shares technology underpinnings with the new QFabric. While switch models and the scalability of switch clusters vary greatly by vendor, a configuration of 1500 ports can be built reliably and simply without the need for fabric technologies.
Action Item: While networking fabric technologies are important for customers to understand, CIOs should first determine if existing technologies can meet their requirements. Virtualization does require a rearchitecture; the move to 10Gb Ethernet is a good time to get the infrastructure groups together to assure that configurations have the flexibility and scalability for a modern data center.
Footnotes: Related Resources:
Data Center Fabric Architectures by Ivan Pepelnjak (IOShints)
Network Fabrics, L2 Multipath and L3 on Wikibon
A Comparison of Current Spanning-Tree Elimination Strategies by Jeremy Filliben
