Networking progress is more than just incremental bandwidth and lower prices

Networking is important–truly important.

“Networking” here isn’t the kind that appears in simple-minded career advice, like, “let acquaintances know what kind of job you want (except when you shouldn’t)”; it’s the deeply technical kind that has to do with IPv6 and spanning trees and expanded roles for SIP.

Too often, though, networking is presented in naive terms such as “software-defined networking (SDN) saves costs” or “Cisco certifications are valuable“. Networking, in fact, is moving to the center of computing in more interesting and deeper ways that deserve clear understanding.

Expand capabilities, not just lower costs

First, SDN is about much more than “cost-saving”, however often it’s presented that way. The real wins with SDN are in manageability, flexibility, and scalability. While these three qualities all have significant financial benefits, they aren’t readily expressed in terms such as “30% cheaper hardware“.

Modern computing needs the flexibility SDN provides because our requirements for networking have become so much more complex. Figures-of-merit for working networks now are a least three-dimensional. Rather than being reducible to a single bandwidth measurement, as was often realistic when e-mail and file-transfers dominated traffic, today’s emphasis on multimedia, cloud, and mobile applications means that network latency and jitter constrain usefulness as much as bandwidth. Moreover, the biggest bandwidth consumers are episodic: they might be videoconferencing this week, but application delivery next week. Priorities among categories sometimes vary independently of load. Only software-defined agility can hope to keep up with this level of complexity.

At a far finer-grained scale, computing needs more performance and manageability from networks. Computational power has increased more rapidly than that of storage, with networking somewhere in the middle. This is why common architectures frequently exhibit five or even more levels of cache: it’s the only way we currently have to keep processors from starving for lack of data.

Fine-grained, rack-level networking experiments

New fabric networks will not only feed data to processors faster; they’ll allow for computing’s components to be disaggregated and “re-factored” into rationalized arrangements. Instead of a “computer” (such as a modern blade) consisting of the contents of a physical box, including CPU (central processing unit), local storage, and interconnects, very high-speed network fabrics will connect scalable collections of CPU with storage. Exceed the capacity of your CPU configuration, and you won’t have to throw all your hardware out; instead, you might upgrade just the CPU array, and manage storage and networking independently. That’s part of Intel’s Project Scorpio rack-level design.

Somewhere between these scales, SDN is also a necessary component of the software-defined datacenter (SDDC) and “responsive” datacenters which automate application-delivery optimization (ADO) and modern application performance management (APM). When you’re planning the future of your networks, make sure you consider all the advantages you can create.