Internet Core Stability Solved – Mode Introduces SD-Core Backbone for SD-WAN

Anyone who wants to minimize the latency of their Internet-based SD-WAN connection should avoid the Internet core. But how do you do that? A global, independent backbone offers an excellent alternative. Until recently, Aryaka and Cato Networks were the only providers of a global, independent backbone. A new player, Mode, just emerged from stealth with significant pedigree and introducing its own independent backbone leveraging a unique routing algorithm to reduce latency.

Internet core: the source of Internet latency

As I noted in our previous post, SASE Experts just completed an in-depth analysis of Internet latency. We evaluated Internet latency and latency fluctuations across the public Internet and a global independent backbone, in this case, Amazon’s network. You can read more about those results here.

What we found was that if you’re concerned about reducing latency, don’t worry about the last mile. Instead, focus on the middle mile, the Internet core. Last-mile performance is actually less stable than the middle mile, but given that the middle mile constitutes more than ~90% of the latency on a global connection, the impact of last-mile latency fluctuations is negligible on the overall experience.

The high latency across the Internet core certainly comes from the long geographic distances between the first and last miles, but that’s not the only factor. The real issue is “Internet distance,” the meandering path often taken by Internet packets due to Internet routing practices. Internet distance significantly exacerbates geographic distance. The congestion at public Internet peering points doesn’t help either, add latency and packet loss.

We examined the impact of replacing the Internet core with an independent backbone — the AWS network. When AWS workloads communicated across the AWS independent backbone (and not the public Internet) latency deviated nearly 90 percent less from the median than when they operated across the public Internet.

Understanding where performance problems emerge in the Internet, lets organizations be more flexible in how they plan their MPLS transitions. Without a global independent backbone, organizations should be rightfully concerned about the latency applications will experience across the Internet. Yes, on any one day, latency across the Internet might be fine, but as IT professionals, we’re concerned with providing a dependable, repeatable experience day-in-and-day-out. The Internet within cities or limited regions can be predictable enough for most applications, but latency across long distances is a different story. A global, independent backbone address this problem.

And since it’s accessed by Internet last-mile services, a global independent backbone lets enterprises choose their last-mile provider, switching as necessary. In this, a global independent backbone differs significantly from carrier offerings, such as those from AT&T and Masergy, who have long provided VPN access to their global services. Those services, though, require subscribing to their last-mile services.

Selecting an independent backbone

Every global, independent backbone has its particular spin. Let’s take a look at them one-by-one:


Aryaka’s SmartCONNECT service replaces legacy MPLS services with SD-WAN and networking services for connecting locations. Smart Access is Aryaka’s remote access service for connecting mobile users. Today, the Aryaka backbone spans 26 PoPs across the EMEA, the Americas, and the Asia-Pacific. Aryaka provides WAN optimization, including data deduplication, as part of its cloud. Enterprises monitor their network through a console that provides application and network visibility. Changes to the network can be made to some extent by the customer; other changes require opening trouble tickets with Aryaka.

Cato Networks

Cato Networks was founded three years ago. The company’s Cato Cloud is a cloud-based SD-WAN that connects and secures all company resources —  fixed locations, cloud resources and mobile users. All send their Internet and WAN traffic across encrypted tunnels to the nearest Cato PoP where Cato software secures and optimizes the traffic. Cato Security Services are a fully managed security stack that includes NGFW, SGW, IPS and, most recently, threat hunting.

The Cato Cloud Network underlying Cato Cloud is a global, geographically distributed, SLA-backed IP network of currently 40 PoPs across the Americas, EMEA and the Asia Pacific. The PoPs form an intelligent overlay built across transit services purchased on multiple tier-1, IP backbones. The PoPs monitor the underlying networks, selecting the optimum path across the Cato Cloud network to the destination PoP for each packet. Cato uses what it calls “multi-segment optimization” to improve performance. Cato offers a self-service management model with customers managing their own SD-WAN instances while Cato maintains the underlying network; managed services are optional.


Fresh off last year’s Google-led funding round, Mode is the newest of the backbone players.  The company provides high-speed cloud access using a cloud-based, “software-defined core” (SD-CORE) network. The SD-CORE network offers IT “affordable private-network reliability and quality of service” across the globe.

Unlike Aryaka and Cato, Mode does not provide an SD-WAN CPE. Instead, the Mode network works with any SD-WAN solution or IP-Sec VPN.. This company’s network spans 21 POPs across the Americas, EMEA and the Asia Pacific built on leased connectivity across a private underlay provided by Ericsson and its global telco partners.

Acceleration comes in the form of the Mode HALO Core routing algorithm, which its founders developed while at Cornell University.  The algorithm eventually was the winner of the AT&T SDN Network Design Challenge. The company founders’ original IEEE paper shows that the mathematically optimal HALO approach is the first fully distributed and autonomous optimal control system for packet-switched networks.

“Mode Core intelligently shifts traffic in milliseconds, dynamically adjusting to network changes and traffic flows,” says Mode co-founder Dr. Nithin Michael and co-author on the HALO research. “Our breakthrough in routing efficiency allows Mode Core to deliver reliability, QoS, and cloud elasticity in a single network. We support user provisioning from any SD-WAN or VPN tunnel to allow enterprises to dynamically modify their desired bandwidth at any time.

Bottom line

The shift away from MPLS to SD-WAN is more than just a move to a less expensive networking service. It’s a change in how enterprises think about their networks. No longer do organizations need to be tied to a single provider for the full network. They can buy local access from different network providers and in so doing, companies gain agility and reduce their costs, something that global backbone providers are all too willing to help out with. Using an SD-Core can assure QoS for sensitive applications worldwide. For detailed findings and to see the full results, register to receive our full report for free here.  Or contact us to test these technologies on your network.

Share this post