Performance analysis: Bursting the myth of MPLS performance

Private MPLS services are supposed to outperform SD-WANs, but as one enterprise recently learned, the internet can outperform even private MPLS services

Performance analysis: Bursting the myth of MPLS performance

Quick question, MPLS or SD-WANs, which would you say offers better performance? For the same bandwidth, we’d assume MPLS. After all, MPLS services are managed for a reason, right? However, as we learned on one recent project, internet performance has improved so much over the years that its performance can rival, even exceed, private WAN services.

The situation

The project was with a manufacturing company that wanted to see if modernizing its MPLS WAN made sense. The alternative would be an SD-WAN or hybrid WAN.

The company’s data center in Los Angeles was connected via MPLS to 17 offices across the U.S., Asia Pacific and Latin America. A mix of applications ran between those sites: SAP, CIFS, mail, VoIP, video and file sharing were the most common.

The assessment

As we were conducting our initial network assessment, we started to see some unusual latency on some of the pings from LA. Check out the MPLS performance table below, and see if you spot the anomaly: 

(Note: Using ICMP pings to measure latency is only a rough guide. The most accurate way is to simulate data plane traffic using a traffic generator for the transit traffic. ICMP is low priority traffic, which can result in slower ICMP response.)

So, did you spot the problem? Look at the variance in the latencies. Typically, a managed network might see latencies changing by 10 or 20 milliseconds, but this company was seeing latency fluctuations by as much as 100 milliseconds on some routes! 

Why the delay?

There are many reasons for varying MPLS latencies on a given route. Routing policies is one issue. What's more, when service providers need coverage in underserved regions, they will sign interconnect agreements with local providers. Depending on the quality and routing policies of those networks, latency may suffer. In the other instances, the bandwidth coming into the region may be insufficient for the traffic levels, causing congestion for all customers during peaks. Sometimes, the carrier network interconnect is not in a direct path—think LA to Dallas to Tijuana, perhaps?

The internet: a better choice? 

Regardless of why MPLS performance was fluctuating, the bigger issue was whether there was a better approach. Enter the internet. Take a look at the how internet performance and MPLS performance compare for Mexico and Costa Rica. As you will see, Costa Rica is pretty consistent between MPLS and internet, but there’s an enormous difference when it comes to Mexico:

In fact, MPLS is a significantly poorer choice in this case when going from LA into Mexico. On average, the internet exhibited 66 percent less latency than MPLS. We didn’t trust those results either, by the way, and repeated the tests only to reach similar findings. The internet was the better choice on this route from a latency perspective. Multiple internet connections would still be needed to match MPLS uptime


So, should the company drop MPLS? Not necessarily. Except for Mexico, the company would see better performance across MPLS in most cases. But what our research did show was the importance of getting beyond the catechism accompanying networking today. 

Whether you're a network engineer concerned about maximizing performance and uptime or a CIO/CFO focused on reducing top-line costs, evaluating your WAN performance is the only way to accurately assess your networking options. Network performance varies too much between services and routes within a service to assume that any one technology will necessarily answer your requirements. 

Copyright © 2016 IDG Communications, Inc.

The 10 most powerful companies in enterprise networking 2022