AT&T's Anschutz: CORD Ties in Open Source, Containers

Service providers are keen to achieve some of the same economies of scale that commodity cloud providers currently enjoy by tapping into Central Office Redesigned as a Data Center (CORD).

Network operators manage thousands of central offices, but they're hamstrung by legacy hardware and high costs. CORD, which comes in various flavors, combines SDN, NFV, commodity hardware and cloud services to create more efficient, agile networks at the edge.

The edge affords an opportunity to customize the network and services for individual customer segments. CORD is a management platform that has a greater need to use containers for customization.

Open networking brings additional flexibility in the use of network resources to make customization possible. In the second installment of this Telco Transformation Q&A, Tom Anschutz, distinguished member of technical staff at AT&T, explains how CORD is designed to use containers and open networking to meet the needs at the edge. In Part 1 of the Q&A, Anschutz spoke about the value of CORD as a management platform. (See AT&T's Anschutz Talks CORD & Containers.)

Telco Transformation: Policy-based management is widely viewed as a robust system for network management. Your impression is that it has been unable to meet expectations. With how would you explain the shortfall in its performance?

Tom Anschutz: In my opinion, policy-based management does have the mechanisms needed for end-to-end network management with its enforcement and decision points and protocols to coordinate. The rub is with sparse implementation across a network, the response to exceptions in policy management create situations when human intervention is needed. With current models of policy management, high costs are incurred when policy control is added to a network element. Thus, they have been used for high value needs at network gateways where high-touch services are needed.

SDN-controlled networks make it possible to have many more elements and interfaces to respond to contingencies and customer needs in a policy-like model. For example, the match-action nature of OpenFlow, mated with its ability to work in a proactive or reactive model is a mechanism very similar to legacy policy control. The greater field of deployment and ability to access all the functions in silicon allow many more situations to be handled by automation without the need for human intervention.

Human intervention does not go completely away with automation -- as it is needed in new situations -- where elements need to be programmed to take decisions without the intervention of skilled manpower. This is not only about failure management, but also about improving the customer experience by reducing the time for new service generation. CORD provides a common infrastructure that can be adapted for a broad range of service needs for segments such as the enterprise, the retail customer, and mobile applications.

Its microservices are easily composed or scripted in different ways to create new services and service logic decisions. And these decisions are more likely to be effective if they are distributed to more vantage points than is the case now.

TT: Speaking of customization, containers are meant for this very purpose especially in an environment with rapidly changing needs. How does CORD take advantage of containers?

TA: Wow! That hits the nail on the head of our plans for 2017. Last year, we wound up taking advantage of the containers and microservices architecture for a virtual service gateway, a scale-out application that uses multiple containers as it needs more resources to serve more customers. The first instance of CORD did orchestration with OpenStack and was set up to use virtual machines [VMs.] But we realized that containers have a faster spin-up time and they have a smaller footprint; they use less memory, disk and CPU. Therefore, we want to be able to pull in containers natively and to manage them with something like Kubernetes. In the existing CORD, containers are placed as groups in VMs, and I look forward to them becoming natively supported in CORD.

TT: What are the key components in CORD that make for carrier-grade open networking especially in the context of expanding use of containers?

TA: CORD is an assemblage of available platforms and software. It is built on the foundations of Openstack for setting up virtual machines and the rest of those capabilities found there. It also takes advantage of an SDN controller in the form of the ONOS. A service orchestration layer at a higher level, XOS, coordinates between OpenStack and ONOS, which is lot more sophisticated than Neutron, an API used by OpenStack for networking. The Neutron API enables Layer 2 connectivity between virtual machines. ONOS adds capabilities for connecting carrier Ethernet VLANs, MPLS, and IPv4 and IPv6. ONOS provides control across the service network including the periphery.

CORD has two instances of ONOS; one that manages the infrastructure, and another that is used as a tenant application. This is done to separate the infrastructure provider from their tenants in the CORD security mechanisms. The infrastructure creates common connectivity, like Neutron does in OpenStack, and the tenant controller manages the broadband access [in Residential-CORD] as a customer networking application.

This is needed to correctly support third-party applications and services. Another controller instance helps to create a fabric. Unlike Neutron, these controls allow the writing of routing and MPLS constructs that Neutron does not do.

In the context of containers, we'd like to prove out the scale-out and microservices orientation for the ONOS SDN controller for coordinated orchestration over distributed hosts in a region. ONOS has a built-in distributed database over multiple hosts and it could lose some of them without disrupting the service. The controller does not operate from a single silo at a centralized location, but works in concert with controls over several hosts while achieving the high availability and high performance needs of carriers.

TA: In the early days, CORD and ECOMP had their separate existence and were not able to talk to each other. Now ECOMP has been open-sourced with its APIs and creates opportunities for accelerating the development of new capabilities for CORD or the augmentation of its current abilities to reduce development costs. The two are complementary in exactly the hierarchy we talked about earlier.

TT: With this background of the technological aspects of CORD, how prepared are vendors and what are the prospects of deployment of CORD?

TA: Since CORD is an assemblage of existing technologies, we can pick and choose its vendors who collectively can achieve the objectives set out for it. Our field trials revealed that some technologies were robust and others were not meeting our expectations and will need more work. So, some pieces of CORD will be consumed before the entire system is utilized.

What we have observed is that the system integrators within the IT industry, who have long experience in end-to-end integration, can come up to speed more easily, while the traditional telecom vendors often find the situation confusing.

ON.Lab, a non-profit, distributes open CORD to system integrators like Radisys, which can find new opportunities for both open and proprietary product development or construct new business models for telcos. Additionally, CORD lowers the barriers to entry for smaller Tier 2 telcos who may be unwilling to invest money in a large engineering staff to do system integration and are more receptive to having system integrators customizing CORD to their environment.

TT: What does CORD do for emerging applications like IoT or streaming media that helps them to make faster progress?

TA: CORD helps to bring the NFV infrastructure as well as storage to the edge where it also aggregates wireline and wireless access resources to use them seamlessly. This is not possible with a centralized NFV infrastructure deployment.

Finally, some IoT applications need low latency and high bandwidth, which is better provided at the edge. CORD does not have significant improvement on streaming media but it will be helpful for emerging applications like virtual reality and augmented reality which operate like a video conference call where high bandwidth and low latency improves their performance.

— Kishore Jethanandani, Contributing Writer, Telco Transformation