Network Virtualization: The Future of the OSI Model
Understanding how network virtualization and the OSI model were envisioned at an earlier time simplifies the road to understanding how they are used today and how they can be improved in the future.
It seems like everybody has their own ideas about networks and network layering frameworks these days.
With the advent of software-defined networking (SDN) and network functions virtualization (NFV), in fact, some experts are on the verge of dismissing the layering concept altogether in favor of a fully integrated stack where forwarding, transport, session management and all the rest are just one big layer, or at best, abstract layers that only the controller needs to worry about.
In the midst of all this is the Open Systems Interconnection (OSI) model, which has been preserved mainly as a theoretical exercise since the internet first converged on the TCP/IP stack back in the 1990s.
Out With the Old?
But is it right to start calling for an entirely new network framework just because traditional physical networks are being replaced by network virtualization?
And is there still a rationale to understand the concept of a 7-layer network architecture when modern approaches call for only three or four, or none at all?
Author and network technologist Russ White raised some eyebrows last fall when he stated bluntly that: “The OSI Model does not accurately describe networks.”
While this may be true, it is fair to say that neither does TCP/IP or the current SDN or NFV constructs. Network layering itself is purely conceptual to begin with, so to say that OSI is wrong because it breaks out operations like presentation and session management into their own layers rather than lumping them into transport or networking is merely a matter of perspective.
White went on to say that OSI was designed to carry the concepts that guided circuit switching over to newer packet-switched networks (PSN). While this allowed packet networks to be layered on top of circuit networks, it was rather unwieldy and largely unnecessary with the advent of middleboxes that assumed the packet-switching responsibilities of individual hosts.
In the end, he argues, OSI may be useful when designing applications, it’s not all that great when designing actual networks.
Hold on, said network architect Ivan Pepelnjak at IP Space.
OSI is far more than a simple layering model. It serves equally well as a functionality framework, a guide to the network stack, or even an actual implementation of the stack, such as LCC on Ethernet or CLNP on a network layer. By viewing various network operations in a nice orderly fashion, you can manage those functions in a reasonable, predictable way.
In this light, OSI serves to explain how networks work even if the final architecture does not adhere strictly to the framework on paper.
Unfortunately, he said, OSI has become something of a religion over the years as network students were grilled on the individual layers and what they did without be told that this is only the way networks can be viewed, not how they must be viewed.
Now we are in the midst of the SND/NFV era where the controller handles all of this and the only thing human operators need concern themselves with is what data they want to move where.
SND, of course, only has three layers — application, control and infrastructure — while NFV describes the three functions of infrastructure, virtual networking and management/orchestration (MANO).
But again, both of these frameworks incorporate the basic operations described in OSI, albeit with different terms and couched within different layers/functions.
It’s fair to say, then, the SDN and NFV don’t diminish the need for OSI any more than TCP/IP did. Rather, it encompasses OSI (and TCP/IP) under a new management and orchestration paradigm to make networking easier and more responsive to the needs of modern users and applications.
As Total Uptime pointed out recently, traditional approaches to networking were rigid and time-consuming. Changes required a lot of mapping, coding, even debugging, and then were fixed on the networking until the need for new changes arose.
SDN and NFV virtualize all of that and funnel management through a centralized controller.
So instead of managing each and every device on the network, operators can simply tell the controller what they require and the controller directs each packet through the most efficient path to its destination.
Functions related to applications, sessions, presentation and even the physical network are still with us, but are not layered the same way.
The best way to look at OSI, TCP/IP and all the other network frameworks out there is to liken them to blueprints to a house. Every house has a foundation, walls, roof, wiring, heat, and the like, but some houses will implement these elements differently from others.
If the very first blueprint had wooden walls and a thatched roof, it still provides valuable guide as to what a basic house should look like even though newer developments like slate roofs and drywall have come into vogue.
All networks provide the same basic function: the get data from one place to another.
Understanding how networks were envisioned at an earlier time simplifies the road to understanding how they are today and how they can be improved in the future.