Virtualization: The Internet of Nothing
I don't think that we understand the profound implications of virtualization, probably because we've focused on two very narrow applications.
What's a "virtual" thing look like? Answer: Nothing. What's real about virtualization? Answer: Nothing. What's the hottest thing in the future of the Internet? Answer: Virtualization. The conclusion here is obvious: We're not heading for an Internet of Things, but an Internet of Nothing.
I don't think that we understand the profound implications of virtualization, probably because we've focused on two very narrow applications. In cloud computing and hypervisor-based servers, virtualization lets us divide a server into component virtual machines. In networking, we can use SDN principles to create virtual networks that run on top of real ones and eliminate restrictions on VLAN size. In both cases, we're making a virtual thing that ends up looking like the real thing we started with--and that's just a teaser for what virtualization can do.
Packet networking, which is what both IP and Ethernet provide, is based on the concept of forwarding tables that blend a knowledge of topology that devices learn from each other, with a similarly-learned knowledge of user connections to the network. The result is a set of forwarding rules that are split among the devices and operate together to move stuff from source to destination. When we talk "SDN" today, we're really talking about different ways of getting that set of forwarding rules, not about different forwarding rules. So we're talking about building the same thing in a different way, and that's no revolution.
Suppose we built a network differently. Suppose that every application lived in its own splendidly isolated subnetwork, connected to nothing except its own components and data. Suppose every user lived in a similar subnet. Now suppose that when we gained access to something, we simply had our subnet connected to the application "something's" subnet. Then we'd have explicit access control. We could even look at sender addresses for security, and we could shut off DDoS attacks by snipping a few forwarding table entries. We could deliver video this way, paid content of any sort. We could also deliver all corporate information this way.
How about mobility? We go through all manner of hassles in mobile networks to follow a cellular network customer as the device roams through multiple cell sites--a whole architecture called "Evolved Packet Core" is dedicated to mobility management. But if SDN lets us control forwarding entries independently, why couldn't a mobile gateway simply alter the forwarding rule for a given customer's traffic to direct it to the cell they're in?
Content delivery? We have processes of DNS spoofing to let us find the best cache for video based on where the viewer happens to be in the network--wireline or wireless. We could assign a fixed address to a given video and then let that address be linked to the closest cache point. Sure it would take a lot of addresses, but what's IPv6 good for if not address space? Anyway, we don't have to do this with all content, just the stuff that's good enough to cache broadly through the network.
The metro network of the future could be a network-as-a-service virtual structure, call it "MetroaaS" if you like. VPNs could become applications-as-a-service ("AaaS"?). In fact, everything that we see today as a combination of an experience (content, application) and delivery (IP, the Internet) could become Experience-as-a-Service. We could connect things explicitly, provide the QoS and security everything needs, and control the structure and its traffic a lot better than we can today. We'd still have the Internet, still have IP addresses, still have switches and routers (virtual, to be sure, but there), but we'd have a completely different notion of how that network works.
There are issues here, but not perhaps the ones you're worried about. Yes, it will take some special software control to make this happen at scale, but we have the technology to distribute data and processing elements in place today to do all of this. The real challenge isn't in how this network works, but how we keep it working.
Imagine yourself as a network operations manager. In the good old days, you saw a broken connection, traced it to a router somewhere, and sent a tech to the right location, the right rack, to replace the box. Now in our Internet of Nothing age, do you tell the tech "Go find that virtual router and pull it for service?" Maybe you need to send a virtual tech? That's actually pretty close to the truth, because in the Internet of Nothing we fix nothings using other nothings--virtual artifacts get virtual-based remediation.
Monitoring and diagnostics are virtual services too. So is management. The Internet of Nothing breaks FCAPS processes into a layer of service-linked virtual fixes that most often substitute one instance of a virtual resource for another, and a layer of fixed-resource fixes that are triggered by a specific indication of a hardware failure.
My point here is that to this point we've been racing for the future with one leg screwed to the ground, i.e., to the present. The revolution of the Internet isn't going to be the adding of millions of M2M gadgets; most of them won't be on the Internet directly anyway. It's going to be the harnessing of virtualization--the making of our ultimate Technical Something (the Internet) out of Nothing.
Follow Tom Nolle on Google+!
Tom Nolle on Google+