Should the Fabric Have a Core? Perhaps for a Little Longer
It seems clear that future network architectures will be built on abstract virtual layers rather than through the hardware-centric approach that most enterprises use today. Even networking vendors like Cisco and Juniper acknowledge the power of software in their bid to build SDN architectures atop proprietary hardware systems.
But it cannot be denied that SDN also lends itself well to flattened fabric architectures, which, if taken to their converged, hyperscale extreme, would place networking components within each compute/storage module. This would enable a highly redundant, eminently configurable and broadly scalable network environment well-suited to Big Data, high-speed transactional workloads and mobile-driven communication.
So it is with a raised eyebrow that I greet new developments in centralized core or even branch/campus switches. After all, if the goal is to foster fabric-based switching for scale-out architectures, why on earth would anyone want to invest in a big core switch that would only serve to funnel traffic through a single point of failure?
Clearly, large, centralized switching platforms must still be in demand, or else the top vendors would be the first ones out the door.But the new releases just keep coming. HP, for example, just came out with a new pair of 5400 switches that emphasize redundancy and scalability, plus numerous configuration possibilities. The top end of the line, the 5412 zl2, provides 12 open module slots for support of up to 96 10 GbE ports or 288 01/100/1000 Mbps ports. It also has four power supply slots along with PoE+ support that delivers up to 30 W per port, a primary factor in the drive for improved redundancy.
The 5400 release is all the more odd because it comes on the heels of HP’s latest addition to the FlexFabric line, which is aimed specifically at providing core support for scalable fabrics. The new 7900 is a "modular core switch" that integrates the virtual overlay with the physical underlay, a move that, a la Cisco, is intended to provide a more resilient fabric that enhances network visibility and provides more rapid configuration for highly dynamic workloads.
Dell, meanwhile, is out with the new Z9500 fabric switch, designed to provide 10/40 GbE aggregation throughout either centralized or "distributed core" architectures, the latter being a uniquely Dell concept that has a switch functioning either as a core or distribution device in advanced scale-out architectures. The Z9500 offers upwards of 1 Tbps of total throughput carried through 132 40 GbE ports and up to 528 10 GbE ports, delivering latencies that are half that of previous switch models.
So is it incongruous for top hardware manufactures to say, "You can have your flattened fabric as long as you build it using the same layered infrastructure that you’re used to"? Technically, yes, but I think there is more to it than that.
The simple fact is that flattened fabrics are not common in the enterprise just yet. So as much as we have come to loathe them, the standard Spanning Tree Protocol (STP) will have to serve for a while longer. But as networking architect Greg Ferro notes, through support for TRILL, IP load balancing and other L2 and L3 ECMP (Equal Cost MultiPath) protocols, a robust core switch can in fact act as a bridge between STP architectures and more advanced leaf-spine topologies. In this way, the enterprise is able to maintain current operations without disruption and gradually transition loads to more flexible and scalable fabric architectures.
So by all means, if the core switch in your data center is showing its age and is due for a replacement, don’t hesitate to swap in a new model. Just make sure that you are building for the future, not shoring up the past.
Photo courtesy of Shutterstock.