There are a number of drivers that favor moving away from the common three and four-tier data center networks with Layer 2 and Layer 3 switching, toward a network based on Layer 2 with just edge and core switches (Get it?… ‘See what we mean by “flat?”).
Drivers for Flat Networks
One of the primary drivers for this change is virtualization. Virtualized environments, and clouds, benefit from the fluidity with which virtual machines can move across the network (and across data centers). Some common virtualization technologies are limited though, to Layer 2 environments within a common VLAN. The presence of a Layer 3 hop ads a great deal of complexity in this regard.
One reason we’ve come to know and love the three and four tier networks common in data centers is that we were forced into this architecture by product capabilities, namely port speed and port density. We’ve learned to build a hierarchy of switching because of the building blocks we had in terms of core switches, access switches, TOR switches, et. al. We must admit though, that driving packets up and down these trees of switches comes with complexity and cost.
New Products and Standards for the Flat Layer 2 Network
Leading vendors, including Cisco and Juniper, are ready to address the flattening of the data center network. New standards including TRILL (Transparent Interconnection of Lots of Links) are emerging in support of this transformation. Increasing port speeds and port densities are other enabling technologies.
‘And the Survey Says…
With the momentum toward virtualized and converged infrastructure, data center networks are surely becoming much flatter things, right? Well, we’re not so sure.
We often conduct polls to gauge the community’s position on many topics that we write about. With respect to this topic, we’ve conducted multiple polls from multiple angles, and response has been decidedly anemic. It’s easy to find interest in most any data center topic that we put out there. In the case of network flattening, the lack of interest is striking. We even began to conduct polls with the option of “I don’t understand the question,” and this choice has gotten nearly as many selections as the other options. For the most part though, there seems to be no interest in the topic, as compared to most any other survey we’ve conducted.
This is one point we’d like to share- that the topic of flattening the data center network is not top of mind in the community as far as we can tell, whether because it’s not yet an issue for many, it’s not of interest, or the options and benefits are not yet recognized. While we have no empirical evidence to support it, we believe the reason for this is that data center network refreshes are not frequent events in the enterprise, and some time must pass before there are a critical mass of these events to bring the dialogue from the data center operator community.
We’d love to hear your thoughts (‘trying one more time) on this topic and encourage you to post here.