Blog: The Internet — "The Good Parts Version" with apologies to William Goldman

by Adam Gervin

Who doesn't love The Princess Bride? I got my first copy at the Dartmouth bookstore outside of summer camp (for those who've been reading this blog, you'll recognize the tie-in to the very first Mode post). No, the movie isn't nearly as good.

It was a tight book. A real page-turner. In fact, this masterpiece by William Goldman was an abridged version of the hideously plotted, unedited, and unreadable original version, penned by S. Morgenstern. Or so Goldman claimed. In fact, it was a clever, but bogus, literary device. Now I digress.

So what do R.O.U.S. and "as you wish" have to do with this next part in our blog series of "Is the Internet good enough for mission-critical business?" Quite a bit, actually. Or at least one bit.

In our last blog, we showed evidence that the Internet was too unpredictable to support mission-critical business applications (or other applications like streaming gaming, and a whole host of next-generation devices, applications, protocols, and services — from IoT to blockchain to 5G).

We must ask: do we dismiss the entire Internet on this basis? Or is there a "good parts version" of the Internet that we can carefully separate, and use with confidence as part of an end-to-end cloud connectivity solution? (I know you were waiting for that tie-in, and yes, you're welcome).

It's common practice to divide the Internet into two parts — the on and off ramps we call Access. The center, we call Core. Access has different names like last mile (or first mile), and Core is sometimes called backbone or middle mile. So can we attribute the unpredictability of end-to-end Internet entirely to either Access or Core? Is it really that easy?

It turns out that we can, for the most part. Internet Access outages are exceedingly rare. Physical network distances at the last-mile are relatively short, and global, long-distance studies show that Access contributions to latency and jitter are relatively insignificant vs. end-to-end values. Solutions like SD-WAN offer additional Access resiliency via redundancy. Additional studies of last-mile U.S. operators reinforce this truth: Internet Access is predictable, can be trusted, and is getting better all the time.

This makes sense when you consider how well-capitalized and well-marketed Access solutions are. It's a highly competitive marketplace, and poor performance is a deal-breaker.

The Internet Core is an entirely different story. It's not nearly as visible, not heavily marketing, and is composed of complex peering arrangements that shift continually. Its "shortest path" priorities are often economy, not performance. Third-party tests show that the vast majority of latency variation (jitter) happens in the Internet Core, as do daily globally-dispersed Core outages that have not gotten better over time.

This leaves us with a remarkable conclusion, one our SD-WAN partners have known for some time: the "good parts version" of the Internet is Internet Access. It is the Internet Core than requires a more predictable, higher-performance alternative.

SD-WAN has led the way by bringing software-based control the edge of the enterprise WAN. What if we adopted this approach, and brought software control to the Core? That would be remarkable. But how?

This isn't a Princess Bride-style "battle of wits." We know the answer, and we'll happily tell you in our next blog installment.

We aren't alone in our search for an Internet Core alternative that's SD-WAN-centric. Steve Garson @ SD-WAN Experts just published an article detailing various ways to take the goodness of SD-WAN and extend it end-to-end across the WAN core. You can read Steve's article here.

Yours truly,