Overlay networks: Understanding the basics, making it a reality
A comprehensive collection of articles, videos and more, hand-picked by our editors
You can all sigh with relief. Software-defined networking (SDN) is officially real because Cisco has finally embraced it. Mark end of sarcasm. Yet Cisco's improved outlook on SDN did get me thinking about network virtualization. Network virtualization is different from SDN. Depending on whom you ask, SDN is either completely different or slightly different. Instead of "just" abstracting the control plane, network virtualization aims to abstract the physical layer of the network from the logical layer. In short, any port on any physical device could provide Layer 2 connectivity for any logical network device in your data center. If this sounds ambitious, then you see why I'm asking the question: "Is network virtualization vaporware?"
Remember Y2K and why it's relevant today
Back in the Wild West days of Y2K, I was tasked with keeping 1,500 Windows 95/98 systems Y2K-compliant. In support of this effort, I reached out to Microsoft to help implement its desktop management system, SMS, to automate patch management.
The concept is exciting, and there are some heavy players that seem to be all in. Still, I can't base architectures on vaporware.
There was one problem. The SMS version at the time didn't support our Novell client/server network staff. Microsoft promised that a release of the product that would solve our compatibility problem was imminent. Let's just say that a few months before the big date, we were still waiting on the software. That was my first experience with vaporware. I had heard of shareware and freeware, but not vaporware. I soon learned it's a product that hasn't been released and doesn't have a shipping date.
I've since become very leery of getting excited about vaporware, and to a much lesser extent will I change long-term architectures based on such software and hardware. But network virtualization is vaporware that tempts me to change my mind. The idea that I can design a network that's completely abstracted from the underlying hardware gives me pause for thought. The ability to "pause, rewind, record" the network gives engineers abilities that would make a seasoned x86 virtualization administrator green with envy.
You might not get the consolidation benefits of x86 virtualization, but the operational capability may surpass that of x86 environments. However, a mature virtual network industry is the catch. I can think of a ton of technical hurdles for the technology. For one, where are the network hypervisors for this technology going to run? How do the physical and virtual network interact with the hypervisor, and what new challenges does this model introduce? X86 virtualization didn't yield performance on par with physical servers until processors and chipsets fully supported and enhanced virtualization.
Still, I'm not holding my breath
That said, I don't see Cisco jumping in all of a sudden to provide the engineering know-how needed to do the same type of hardware optimization in network gear as there has been in x86 servers. That leaves projects such as the solutions Intel and others are spearheading. But are these open switch-based approaches robust enough to provide the same level of performance of, say, a standard 96-port 1 GB switch compared with a 96-port virtual switch spread across 20 physical devices connected over large distances?
I'm sure you can think of other challenges. But the point is that this is a tough problem that needs a large, combined industry effort to tackle. VMware seems to want to lead the way with its network hypervisor NSX. The potential is great, but it is, to date, still vaporware. And that's my point. The concept is exciting, and there are some heavy players that seem to be all in. Still, I can't base architectures on vaporware.
What are your thoughts? Will virtualized networks stay the things of Disney movies, or is it a technology that will materialize sooner rather than never?