Subscribe

MPLS is dead. Long live MPLS

How do companies go from multiprotocol label switching to an SD WAN network?

Stuart Hardy
By Stuart Hardy, business development director of EOH Global Networks Division UK.
Johannesburg, 12 Jul 2016

If you are like me, then you believe there is a change coming that will forever alter the way networks are seen and used, especially multiprotocol label switching (MPLS). The basis of this belief is that at some point in the near future, users will consume all business applications from some cloud infrastructure or software as a service platform. And in this world, traditional MPLS networks are complex, expensive and inflexible. Dead.

And the technology that will save the day, in much the same way virtualisation saved servers, will be software-defined WANs (SD WANs). SD WANs are widely understood as the ability to decouple the forwarding network (data plane) from the routing and decision-making (control plane). Sounds simple enough, but it provides a significant amount of opportunity that allows companies to resolve some of the key challenges arising from the evolution of cloud IaaS, SaaS and PaaS infrastructures.

But, how does a business go from a current network architecture like MPLS to an SD WAN network? That's what a lot of companies are struggling with today.

In global networks, where applications are being delivered over long distances, the results of using the Internet for application delivery are poor. The main reason is because in global networks, Internet latency is high and packet loss exacerbates the issue by reducing bandwidth even further. The effect of latency can be radically improved by technology such as TCP IP acceleration, which resolves the TCP window size issue, but is more effective in an MPLS network than the Internet, because MPLS has lower latency and no packet loss.

Traditional content delivery networks (CDNs) can't solve application delivery either. CDNs use the lowest and cheapest form of Internet to synchronise their caches. They have high drop probability, high latency and high packet loss. This is the same bandwidth that would be used to deliver business-critical real-time applications. CDNs are very good at delivering static content like Web traffic, but real-time applications are dynamic, and as a result, need real-time network performance. To resolve this issue, a company would have to put in dedicated L2 or MPLS bandwidth, which would undermine the economics of the CDN business model.

Problem with MPLS

One of the main characteristics of MPLS is that it backhauls Internet through the network. That has been fine, as long as the Internet has remained relatively small and the applications coming from the Internet have not been business-critical. But, that's changing, and more information and business-critical applications are being delivered from the Internet.

Most companies are still trying to deliver these applications through backhauling Internet through MPLS, which is increasing costs and further taxing application performance as more latency is introduced. There is also no answer to this problem in traditional MPLS network design and technology, as it is not designed to co-exist with the Internet.

Traditional content delivery networks can't solve application delivery either.

To address this problem, companies are starting to use software to manage networks and application delivery. This is the first step of SD WAN, or what is referred to as a layered network approach. Even though, in many cases, these are still deployed as hardware solutions, products like Riverbed, Ipanema, Silver Peak, etc, have evolved their technology from just MPLS application management to support using the Internet network in association with MPLS to get the best of both worlds. Now software is being used on top of the physical network to make the decisions for application delivery across the network.

Spoiled for choice

For a long time, this technology has only been available to big-budget businesses, but with a huge increase in competition from companies like Ocedo, VeloCloud, LineFactory, Silver Peak and more, companies have lots of choice when it comes to SDN solutions as an overlay to MPLS and Internet.

However, what is evident in these network designs is a drop-off in user experience for applications delivered from the Internet, like 365 and SalesForce.com, due to the already discussed high latency and packet loss evident in the Internet. And it's not everywhere in the network, it's most likely only in areas of a global network, where the application is in excess of 100ms away from the user.

The optimisation capability found in a lot of these SD WAN solutions cannot provide the same level of application performance as they are able to in MPLS networks, as the Internet is less predictable.

This is pushing customers to roll back some cloud deployments and investigate architectures like Express Route and Direct Connect. MPLS, ultimately, will resolve some of the issues, but the whole basis of using MPLS to get to cloud undermines the principle of what real cloud represents. For many companies, this is frustrating as the cost of deploying these MPLS extensions in most cases is expensive and reduces the scope of leveraging cloud services.

There are solutions. In my next Industry Insight, I will look at the network of the future, and why I believe SD WAN will replace traditional networking.

Share