Howdy folks, Greg is out on vacation (adventuring in the magical kingdom of Orlando, Florida) so I will be filling in for a bit to continue our series on “wargaming” WAN scenarios.
In previous posts, we established our hypothetical network and looked at how SD-WAN implementation can affect bandwidth and total cost. I encourage you to go back and start reading from the beginning. I can wait.
Today, we’re going to continue down the SD-WAN path and explore a scenario some WAN managers have devised for keeping MPLS only at major sites, which we call the “MPLS-core” network.
Why Keep MPLS at Core Sites, If Not Elsewhere?
While many WAN managers want to move more corporate traffic to the public internet, there are some MPLS features they don’t want to leave behind. Understandably, the class of service and security guarantees of running on a private network seem crucial for their largest sites, which have thousands of employees and bandwidth requirements up to 1 Gbps.
While many WAN managers want to move more corporate traffic to the public internet, there are some MPLS features they don’t want to leave behind.
At the same time, there are a variety of reasons—which we detailed in our MPLS-DIA network model—that enterprises want to make the move to local internet breakouts.
So what’s a network manager to do?
Several of the WAN managers we interviewed for our 2018 WAN Manager Survey had all developed a similar solution. They planned to keep MPLS, but only at headquarters and data center sites, going full internet with the rest of their WAN.
Often, they reached that decision because their networks were comprised of many small offices or sites that, as one interviewee from a financial company said, “want to connect to the internet more than the intranet.” After adopting SD-WAN, only a handful of this company’s largest sites will remain on a private MPLS WAN.
Another interviewee with more than 3,000 sites indicated that they were eventually ditching MPLS at all but 400 crucial sites. Because of this move, they were already seeing savings of 25% in developed markets and up to 50% in developing ones. They were willing to “take a little risk with [their] smaller sites” and found that, generally, their end-users “don’t even realize the difference [of DIA compared to the] MPLS they had before.”
Let’s see how it works in practice when we reduce our MPLS footprint to only a few large sites.
Building Our MPLS-Core Network
To model the MPLS-core, our hypothetical scenario kept MPLS at only the 13 sites where MPLS ports were >100 Mbps. All other sites are a DIA-broadband hybrid.
We generally kept a DIA port that is at least the size of the original MPLS port, as well as a matching (or larger) broadband circuit. We included 35 sites with broadband circuits in the 51-100 Mbps range.
How Does Keeping MPLS Only in the Core Change Our WAN?
Original MPLS WAN Site Capacity Range
MPLS Core DIA-Broadband Hybrid WAN Site Capacity Range
The resulting network has a global average of 420 Mbps available per site, which is a significant increase over the original 125 Mbps. Again, this average is pulled up by the regions where large offices and data centers are located that include very large ports and circuits.
How Does Average Site Capacity Change by Subregion?
MPLS Core-DIA Hybrid WANs Average Total Site Capacity by Subregion
Much of the net capacity increase was in the U.S. & Canada, Western Europe, and East Asia, where the majority of larger sites are located.
The added capacity in other regions, while on a smaller scale, is nothing to sneeze at either. Excluding Sub Saharan Africa, all other regions saw significant increases to site capacity (some more than doubling on average) compared to the original MPLS network.
But How Much Will It Cost?
As SD-WAN prices are based on the amount of encrypted throughput at each office, as we increase the active bandwidth available at each site, our SD-WAN costs will also increase.
Our unmanaged SD-WAN costs went up from about $450,000 in our previous MPLS-DIA hybrid network to about $710,000 for the MPLS-core network. But, despite that increase in SD-WAN costs, we see a 45% discount compared to the original MPLS network.
How Did Keeping MPLS Only at Core Sites Affect the TCO?
Original MPLS, MPLS-DIA No Backups, & On-net DIA No Backups with Unmanaged SD-WAN Annual TCOs
Original MPLS & MPLS Core On-net DIA-Broadband WANs with Unmanaged SD-WAN Annual TCOs
Again, keep in mind that as our DIA is on-net, you would have to be fortunate enough to have a mostly urban network largely located in multi-tenant office buildings or the like. However, even with some additional access loops for DIA ports, this configuration would likely be a significant discount over the original network.
The hybrid WAN with MPLS only at core sites boosted capacity more than 200% and saved 45% compared to an all-MPLS network. In summary, the MPLS-core scenario is attractive to WAN managers willing to move most of their network traffic to the public internet but who can’t say goodbye to MPLS entirely.
Moving Beyond the Core
That's not the full story though. Most of the WAN managers we spoke to consider this model a halfway point to phasing out MPLS altogether. “We are moving away from MPLS. If possible we will be moving toward getting rid of all MPLS,” one interviewee said. “Most of my sites are very small—there are around 400 large industrial sites that need connections—but the future is SD-WAN.”
Most of the WAN managers we spoke to consider this model a halfway point to phasing out MPLS altogether.
Another WAN manager in the financial industry mentioned that they are considering reverting back to private line connections between their corporate headquarters and data centers as an eventual alternative to MPLS.
They explained that, “the primary strategy is that we expect the core of our sites, which are offices, to transition to an SD-WAN solution and only the largest sites to remain on a private WAN. For some, that will be MPLS, but other radical thinkers (like me) think we can transition to point-to-point Ethernet connections between them. We will connect them to each other and an Equinix facility; a return to a hub and spoke design but with a smaller number of spokes.”
Definitely an intriguing prospect.
Join us next time as we take the plunge into modeling all-internet WANs to see what a post-MPLS network could really look like: