Blog: Second Floor! Internet Core Measurements and Office Supplies

by Adam Gervin

As I sit here sipping my fourth Coke Zero of the day (true), next to a former starting outside linebacker for the 49ers (also true, and a much longer story), I am reflecting on how to break some very exciting news to you.

That's it! I’ve settled on the direct approach:

Steve Garson, of SD-WAN Experts, has agreed to be our featured guest for the next Mode Webinar. No take backs!

And the topic ... it's Gold, Jerry! P-u-r-e Gold!

**SD-WAN: Three Must-Do's for Peak Performance**

Could it have to do with your SD-WAN? Well, yeah. Does it talk about MPLS costs and inflexibility? Maybe. End-to-end encryption, and enterprise control of security policy? Very possibly. Extending edge control to the network core? Definitely.

Featuring: Nithin Michael, Founder & CTO, Mode;
and Steve Garson, Founder, SD-WAN Experts
When: Wednesday, November 28th, 10:00 a.m. (PDT)
Register Now

Now that that's out of the way, we can resume our walkthrough of Mr. Garson's excellent keynote presentation at the ONUG Fall event in NYC. If you recall, his talk offered three experiments relating to Internet Core performance. Last week we covered the first, in which Internet Core response time was extrapolated from measurements of long-haul response vs. last-mile response.

In the second stage of the presentation, he used Catchpoint to directly measure Internet Core response time among the same collection of endpoints, comparing them to the first data set.

The goal was to validate the original data by using two distinct measurement methodologies, and hopefully get similar results.

And it worked! Let's look at the following raw data table:

It's pretty clear that the measurements from Part I (last week) and Part II (this week, direct Internet Core measurement) are practically identical (within the margin of error). Both the median and the standard deviation.

With one exception — the route between Melbourne and San Jose AWS. It's likely that Telstra (the backbone provider) is doing something special in moving data back and forth with San Jose AWS. This would account for this path's resistance to latency variance (when compared to generic Cedexis data obtained across a collection of providers, as used in Part I tests).

This leaves us with two takeaways:

  1. Part 1 data is independently confirmed via a second methodology. Again, the overwhelming contributor (>99%) to long-haul response variance is the Internet Core.
  2. If you optimize this long-haul path somehow — we like to call it Core Control — you can get long-haul performance that acts like last-mile performance. Wow. We might have something to say about that soon :) The devil, as always, is in the details.

Next week, come back to for analysis of Part III of Steve Garson's research project and presentation. But for now Save the Date, November 28, 2018, as we host Mr. Garson for a not-to-be-missed Webinar, live from Silicon Valley and the Mile High City. So exert some control — over your schedule — and pencil us in. Your SD-WAN can't afford to miss it.