Cloud Strategies

Are Overlays the Answer for Telco Data Centers?

Telco data center operators have recently begun to consider how to bring the benefits of virtualization to the networking domain. OpenStack has been rapidly gaining traction, but concerns remain that it lacks some of the functionality required to support highly available and highly scalable networks. Numerous vendors have taken up the challenge of filling that gap.

In my recently published report, "Overlays for Data Center Networking: Drivers & Challenges," I explore some of the expected benefits and perceived challenges surrounding overlay networking. It highlights and analyzes the potential impact overlay networking will have on the telco data center networking domain, including the various drivers that support its deployment. It also provides analysis of some of the main challenges that overlay networking can introduce into the environment. It also analyzes the strategies of ten key overlay networking technology suppliers.

Overlays have been proposed as a way to provide the connectivity needed to support the networking needs within the data center. They abstract the details of the physical network, making it much faster to connect virtual machines (VMs) and other devices. Rather than provision paths on physical devices, overlays encapsulate traffic using protocols such as VXLAN and NVGRE to tunnel across the physical network. These newer protocols allow operators to move beyond the limitations of VLANs, which only support 4,096 virtual networks, so they can better support multi-tenant cloud services.

Another driver, which is not as clearly understood, is the emergence of network functions virtualization (NFV). In NFV, functions that had previously resided on purpose-built, proprietary platforms will be supported on general-purpose servers in the form of virtualized network functions (VNFs). How these workloads will behave in terms of elasticity and mobility is still being determined, but overlays are seen as an effective approach to providing the flexibility they will need to manage VNFs. Overlay networks also make it easier to move workloads between and across data centers. By mapping VXLAN to MPLS paths, virtual private networks can be extended beyond the data center across the WAN.

Despite these potential benefits, overlays are often seen as introducing more complexity into the data center environment. Opinions vary as to how significant this issue is: Some argue that in the end, operational expense will decline, because less time will be spent managing the physical network; others contend that managing service performance will become more complicated from having two domains to consider. Even when overlays are present, the physical network will remain critical for ensuring service quality. Discussions with operators suggest that the benefits will ultimately outweigh these concerns, although it will take time for operations to get comfortable with the new approach.

— Roz Roseboro, Senior Analyst, Heavy Reading

DHagar 3/17/2015 | 1:37:59 PM
Re: Large data centers ditch overlays Dredgie, makes sense.   It sounds as if this can be a missing link to putting an effective "system" together for the Telco Data Centers.  That's good news.
Dredgie 3/17/2015 | 1:12:09 PM
Re: Large data centers ditch overlays > Absolutely - you can run hybrid. Naturally open source is the key, here. Call it defacto standardization, to be fair. If you get someone like redhat in the loop (as in Calico) and are in their (RHEL) distro's, then it becomes easier for the mid-size players to adopt. But L3 isn't for everyone. If I'm a small DC, then I could stick with L2 as I won't hit the scale where such solutions become overly complex. 
DHagar 3/17/2015 | 12:57:15 PM
Re: Large data centers ditch overlays Dredgie, sounds good.  This could then become a viable "utility" to provide the overlays needed?  Note:  As long as there are multiple major providers to fill this space, and/or it's enough of an industry standard that it drives the smaller providers as well?
Dredgie 3/17/2015 | 11:58:10 AM
Re: Large data centers ditch overlays > Far better scalability. It's Internet scale (BGP routing). If you look at the big data centers today (think google, FB, Amazon, and the like), none are using L2 overlay techniques. They are all L3 of some kind – just not standardized. You don't have to be as big as those guys to look at layer 2, though. Not even close.

L3 is also easier than L2, once the right hooks are in place: rip/replace OVS with the Linux router. Add a BGP stack then some back-end ACL and cloud OS plug-ins. RR for scale and you have the basics in place. No overlays. Not encap overheads. 
DHagar 3/16/2015 | 9:17:11 PM
Re: Large data centers ditch overlays Dredgie, interesting option.  Does that provide the same level of scability?  If so, it could become a viable alternative, as you suggest.

I am in agreement with Roz, it appears that the overlays will provide the foundation to truly solve the growing complexity and serve the growing demands of data centers.
Dredgie 3/16/2015 | 7:41:20 PM
Large data centers ditch overlays > There is an option to ditch the overlays partially or entirely and opt for all-IP with open source initiatives like Project Calico. (http://www.projectcalico.org) Even the OVS guys are getting into the game with OVN. I believe this is becoming the preferred deployment model for LXC abstractions like Docker in larger data centers and makes those MPLS interconnects easy.
Sign In