Fool's Gold

by Adam Gervin

The Gold Standard Needs Polish

Everyone reading this blog has lived long enough to see records broken. Giants laid low. Things thought impossible, made commonplace. Gold standards surpassed and forgotten. In the modern era of business connectivity, MPLS has been the gold standard for quite some time.

A real 90's throwback that everyone knows has stayed around a bit too long, like melted Vanilla Ice. And yet, MPLS use is growing, year over year. The truth: exploding bandwidth, combined with inertia, is camouflaging an otherwise declining tech.

The stake through the heart of MPLS will be alternative solutions that are faster, better, cheaper. SD-WAN is the tip of this sharp instrument, and SD-CORE the follow through. It won't happen overnight. It will be a gradual realization of parity, followed by a soft coup. It's already well on its way.

It began with this question: is MPLS really as good as it gets?

There are already whispers among CIOs about things like underwhelming MPLS performance, or lapses in virtualized tenant separation, hence privacy. But leave it to Steve Garson and the folks at SD-WAN Experts to ask this blunt question: if an SD-WAN and an MPLS connection have the same bandwidth, which offers better performance?

To answer this, they examined a manufacturer considering a switch from MPLS to SD-WAN, starting with a branch in Latin America.

On the MPLS side, they found occasional routes with latency variations up to one-tenth of a second, inferior performance indeed. This could be due to policy. Or it could be related to the fact that no service provider has a true global MPLS network — it's assembled through business relationships with countless regional players, often with unpredictable results, perhaps including this one.

SD-WAN Experts then examined specific routes between the US and either Costa Rica or Mexico. In the case of Mexico, MPLS scores were significantly poorer than Internet. In fact, on these routes, at that point in time, the Internet had 66% less latency variation than MPLS.

Of course, this isn't usually the case. MPLS is generally reliably good (so says the SLA and the nosebleed pricing), and the Internet is ... unpredictably good (or unpredictably bad). Or simply, unpredictable.

One thing is clear: you need network redundancy to deliver true high availability. Even MPLS, on its lonesome, is no longer a safe bet.

Let's consider the mechanisms of redundancy. SD-WAN offers last-mile redundancy. That's a good start if you use it.

When it comes to the core, the wise company also seeks redundancy. For example, rather than backup MPLS with another MPLS connection, consider using an SD-WAN with a high availability software-defined backbone. It could make all the difference, and save you a lot of money to boot.

Many Internet Core alternatives have different flavors of redundancy baked right into their designs. Optimized Internet approaches use POPs to probe various opaque public IP paths, slowly switching among them based on performance guesstimates. A sort of redundancy, albeit limited and crude.

The ultimate form of redundancy would be to roll up every network on earth under a perfect, autonomous control overlay. Not only would you benefit from the entire pan-global capacity, but you'd also have software scouring all those connections for the one, optimal path for your traffic, every 150ms.

Mode does not control every network. Yet. :) But it does have a perfect autonomous control overlay. Mode SD-CORE capacity starts with a global underlay provided by Ericsson and nearly 100 service providers, and it grows with each SP underlay we add. No limits. An autonomously controlled network with massive capacity offers the ultimate form of redundancy — the fastest possible combination of measure, calculate, and control, routing packets around any obstacle, tripling effective network utilization. Finding the right way, at just the right moment, considering all available options.

MPLS may not be the universal gold standard we thought it was. That's ok, we can all live with that. It's worth it, should it help us recognize the peril of putting all our eggs into one basket.

It's time to diversify. Go ahead and backup that MPLS! Redundancy is our friend, and SD-CORE is a big step towards a No-Worry Network.

You can read Steve Garson's original piece here. Make sure to come back next week for another take on the need for redundancy!