SDN Network Models
#SDN models: switches and overlays
As with most new exciting (read: hyped) technology there's always some initial fragmentation that occurs in the market. Everyone wants to have their fingers in the newest pie and from that comes what musicians call "variations on a theme." The melody is the same, but the harmony and chords are enough different so as to make them appear to be different songs.
SDN is no different, and there is already some fragmentation in the basic SDN model. While ONF continues to focus on the "original" switch-based model, offerings that instead use an overlay-model are beginning to get noticed.
Not to belabor one of my favorite quotes this year but...
"If you look at the standard SDN model, [Layer 4-7 services] are applications that can basically run on the [SDN] controller platform. But that's not the only way to do them. We'll hear about different approaches. Network services for SDN are going to be a big story in 2013."
-- Brad Casemore, "Networking outlook: Controllers, Layer 4-7 will roil SDN 2013 market" [emphasis mine]
It's not just the L4-7 services that can be done "differently" then the first incarnation of SDN. It's L2-3, too, that will ultimately fragment into 2 or 3 core models, each designed to solve the same problem but having different architectural and business benefits.
For example, the switch-based SDN model relies heavily on standardization and commoditization in the network. Today that means OpenFlow-enabled switching fabrics running on commoditized hardware (note that this overlaps well with the notion of network virtualization in general, because the OpenFlow-enabled switches could be virtual or physical, depending on desire and need).
This model, however, is somewhat disruptive and capital intensive, because it requires investment in the switching fabric. That may mean upgrades to firmware or new hardware. Either way, it's disruptive. The benefits of the disruption include agility, lower operational overhead than is incurred by managing your network node by node, and a network that is ostensibly more resilient and able to adapt without human intervention.
The overlay model, on the other hand, benefits primarily from being non-disruptive. It assumes an existing, L3 IP fabric atop which it lays its own, virtual network using tunneling methodologies. Solutions like Midokura's MidoNet, VMware's Nicira and VXLAN, and Microsoft's NVGRE are overlay-based SDN models that seek to minimize disruption while enabling a physically agnostic network topology that is better suited for cloud and agile infrastructure. The flip side, of course, is that you're now managing two completely different networks, which is necessarily going to have an impact on operational overhead.
Neither of the models is perfect, and I'm sure someone from each "side" would argue the other isn't really SDN. But the notion behind SDN is to address some very real problems around network rigidity and reliance on fixed IP network strategies. Both do that, just in different ways.