The software defined operator

Any prediction remains difficult, especially if it is about the future [1]. But, predicting that Internet traffic will continue to grow exponentially seems to be a safe bet. We at Deutsche Telekom are committed to deliver the Best Network to our customers. In order to meet customer and investor expectations with the traffic demands of the future, network operators need to act - the old way of building networks and producing services will hardly scale for the future.
Google, Facebook and other Internet companies have proven that a new software and datacenter-based production model can deliver high flexibility, agility and highest availability. By latest with the move towards an "All-IP" network, network operators have to implement a similar approach in their production chain in order to be competitive. At Deutsche Telekom, we have summarized this strategy under the name Software Defined Operator.
For us, the Software Defined Operator has three main elements - the drastically simplified IP Network, the Infrastructure Cloud and a SDN-inspired Realtime Network and Service Management.

 

Drastically Simplified IP Network

Since the first carrier offers in the 1990s, our IP networks have become rather complex. Most operators are running a two-tiered IP/MPLS core on top of an OTN (Optical Transport Network) and a rather complex aggregation for broadband customers, including the termination of customer PPP over Ethernet connections on BRAS (Broadband Remote Access Servers). With the continued growth of IP traffic and the move to "All-IP", we have to ask ourselves the question whether we still need all the complexity to deliver our services.
At Deutsche Telekom, we are driving the simplification with the current implementation of a Broadband Network Gateway into our production networks (e.g. in Germany and Greece). In parallel, we are developing a revolutionary long-term approach under the name TeraStream. We are currently running limited TeraStream customer pilots in two countries.
TeraStream only has two types of routers - R1 as customer facing edge router and R2 as Datacenter and Peering facing edge router. Optical transmission is fully integrated into the IP routers using 100Gb/s Coherent interfaces and an extremely simple optical drop and waste concept. TeraStream does not use any OTN - the only optical components in the network are amplifiers and very cheap optical splitters. TeraStream is agnostic to access technologies, covering both fixed and mobile access.
R1s are IPv6-only routers, IPv4 is delivered as a service - produced in the datacenters attached to the R2s, encapsulated using the standardized Lightweight 4over6 technology [2] and terminated on the customer premise equipment.
TeraStream does not use any MPLS. The efficient IP+Optical approach allows us to overprovision the core. QoS is applied at the Edge towards the customer.
To simplify the routers to the bare minimum, we have tightly integrated network I/O optimized datacenters with the IP network. The operation of these datacenters at the R2 locations follows the Cloud paradigm, we have called this the Infrastructure Cloud.

 

Figura 1 - The Software Defined Operator

Infrastructure Cloud

Traditionally, datacenters are optimized for compute performance and storage access. An ERP application like SAP in a traditional 3-tiered architecture is probably a good example - network traffic in a datacenter supporting these applications typically flows in east-west direction. High availability in such a datacenter is realized through redundancy on the network, server and storage layer.
These more traditional IT-orientated datacenters are not optimal for producing VNFs (virtualized network functions) in a telco environment. For VNFs, traffic typically flows north-south, from the server in the datacenter directly towards the network. Our Infrastructure Cloud model represented the first truly network I/O optimized approach in the industry. In our first pilots, launched in December 2012 in Croatia [3], each server was connected using 6 10 Gb/s Ethernet connections directly to the R2 routers. Most IT-optimized datacenters have this as a total capacity towards the network. High availability in the Infrastructure Cloud is achieved using georedundancy and a tight integration into the IP routing system.

 

Figura 2 - The TeraStream Architecture

We wanted to avoid a single vendor dependency in this critical area. The Infrastructure Cloud is fully based on Open Source. We are using KVM as Hypervisor, Openstack as Cloud Orchestration Framework and CEPH for our software defined storage approach.
Figure 3 shows a view on the Infrastructure Cloud from a very high flight level. The blue boxes represent network functions required to run the network like DNS, DHCP, virtualized routers for VPN services, the Lightweight 4over6 Softwire solution or for a migration a virtualized BRAS.  The magenta boxes are representing traditional core services for fixed and mobile networks, including a packet core, IMS or video distribution systems. The yellow boxes represent options for future partnerships, e.g. on the content or application side.

 

Figura 3 - The Infrastructure Cloud - 40000 feet flight level

The Infrastructure Cloud fully follows cloud paradigms, including a Continuous Development / Continuous Integration / Automated testing cycle. This move to a Telco devops model still requires some focus - also on the supplier side.
The move to an Infrastructure Cloud based service production allows centralization of those services which can be centralized, and it allows to distribute what needs to be distributed. The move to virtualized or, better, cloudified network functions produced in the Infrastructure Cloud represents a major transformation for the telco industry. This is impacting the suppliers who need to partially rearchitect their network functions to make them cloud-ready, but it is also impacting our teams on the operator side. Innovation, Collaboration, and Empowerment of the teams are required to make this approach successful.
The ETSI ISG on NFV [4] had developed the reference architecture shown in Figure 4 - the major challenge in todays implementations is on the orchestration side. While OpenStack [5] is generally accepted as Infrastructure Manager for a Virtual Machine based environment, new developments like the lighter virtualization approach using containers has to be integrated. On the overall orchestration, we are seeing many parallel efforts, both on commercial products as well as on Open Source - but, none of the approached currently meets all the operator requirements.

 

Figura 4 - ETSI NFV Reference Architecture

Overall, Open Interfaces and Open Standards are crucial for the success of the Infrastructure Cloud and the Cloudification of Network Functions. The role of traditional standards is in many areas replaced or at least complemented through Open Source. A word of warning: many vendors are using the word Open when they publish a specification - but, is a specification owned by one entity really open, or is it just "published"? The author strongly believes that Open approaches require a community or standardization organization behind.
The initial version of the Infrastructure Cloud is built using standard x86 server hardware. With a massive rollout, efficiency has to be a main focus. On Jan 27, 2016 a group of operators including Deutsche Telekom announced to join the OCP (Open Compute Project) [6], forming a TelCo working group within OCP.
OCP was originally established with the help of Facebook. Applying the Open Source spirit to hardware, open specifications developed in a community effort have revolutionized the datacenter, leading to a significantly lower Capex and Opex. OCP has developed optimized racks, servers, switches and storage systems.
The OCP approach was recently complemented through the TIP (Telecom Infrastructure Project) [7], applying the same Open Source mindset to the overall Telco infrastructure, including Fixed and Mobile Access, Aggregation, IP Core. We strongly believe that TIP can do to those areas what OCP did to the datacenter. Community-driven open specifications will lead to disruptive innovation across all network areas.
Within TIP, our team from Deutsche Telekom will initially focus on IP-based access, aggregation, IP and Optical integration - also bringing in some related work we are driving, for example on home networking.

 

Realtime Network and Service Management

The author strongly believes that all the simplification in the IP network, all the work around the Infrastructure Cloud is of limited value unless we are changing the way we are managing our networks and services. As part of TeraStream, we have developed a new realtime approach - this was inspired by some work around SDN.
We do not believe in rearchitecting the network or operational support systems for new services we want to integrate - we should be able to "program" service and fully automate the operation.
Our Realtime Network and Service Management approach (RT-NSM) is fully based on YANG, a generic schema description language standardized within the IETF (Internet Engineering taskforce).
Southbound, element managers or specific integration into OSS systems have been replaced by using standardized YANG datamodels to access the physical or virtualized network elements, NETCONF is used for the communication.
But, YANG datamodels can also be used northbound towards the business support systems. Describing services in YANG leads to a fully automated YANG-based RT-NSM, addressing both the traditional fulfillment as well as the assurance space.

 

Figura 5 - YANG-based Realtime Network and Service Management

Summary

Software Defined Networks and the Cloudification of Network Functions sure have the potential to fundamentally change the telco landscape. Sure, telcos won't become Google II or Facebook II - but the technologies originally developed by the Internet companies are adopted by the network operators, both on the software as well as the hardware side. The Open Compute Project has changed the way large datacenters are build, the Telecom Infrastructure Project will change the way telco networks are build.
The broad adoption of the technologies described in this article will happen over the next years. Until then, a lot of focus has to go on orchestration and automation of all services, as well as the scalability and efficiency of the technologies.

 

comments powered by Disqus