Disaggregation is a somewhat clunky term for a key concept across the telecommunications industry for cable, telco and mobile providers. In order to better define disaggregation, FierceTelecom Blitz Week is hosting a panel at 2 p.m. ET on Wednesday.
Disaggregation, the splitting of network elements into smaller parts, helps service providers break down their vertical service silos and enable the use of cheaper white box devices that work with other vendors' software.
While disaggregation mainly focuses on separating the hardware from the network operating system (NOS) and then putting the NOS on white boxes, it can also include disaggregating the software itself in order to only deploy the software that's needed on a device. In the latter instance, service providers are only paying for the pieces or elements of the software that they actually need.
"The way that I see it, disaggregation is a big word, and it's being used way too often these days to describe so many different things," DriveNets' Run Almog, head of product strategy, who is one of the panelists on Wednesday's disaggregation and virtualization panel. Almog breaks disaggregation down into three tiers with the first being the separation of hardware from software that dates back to Google implementing it in its own data centers roughly a decade ago.
"Essentially, it's a single device. It's a single unit," he said. " It's not a multi-chip device, because multi-chip brought a different level of complexity to this scenario.
"The second tier of disaggregation is about separation of the hardware itself.The idea of good disaggregation is to make all of these behave as if they are a single entity. So the outside world thinks that it's a single element or a single router, or a single other network device. But the world outside this cluster of elements doesn't really understand that it's clustered in front of him. He thinks it’s a single device or single entity."
Tier three is the disaggregation of the of the operating system itself by building it in a way that it's modular elements, or functions, within the operating system can act independently from other functions that are containerized.
"So if I have this disggregated chassis, and I have dedicated white boxes for my line cards and fabrics and my controllers, then my BGP container can run anywhere I want within this list of options," he said. "So that disaggregation, the separation of the of the software itself, is what I call the third tier of disaggregation."
A key element of disaggregation in general is the separation of the control plane and data plane with an orchestration layer, or hypervisor, on top to address the allocation of resources, such as cloud, on demand.
After working with AT&T for over four years, DriveNets was finally able to confirm in September that it was providing the core-networking routing software for the telco's next-gen core network. Almog expects more deployments will become public this year.
Over a year ago, AT&T put its specifications for its distributed disaggregated chassis (DDC) white box architecture into the Open Compute Project (OCP.) On the same day, DriveNets announced its Network Cloud routing software was first on the market to support the DDC model.
Almog said DriveNets is currently working with the OCP on contributing simplified interfaces that decouple the firmware upgrades from the NOS.
"The purpose is to decouple the dependency between the NOS and the ODM so that our NOS, for example, can run with any ODM seamlessly," Almog said. "So the purpose is to make this interface as lightweight and as independent as possible so that more players can join the OCP wagon, and not be dependent on other vendors."
CableLabs tunes up FMA, Adrenaline for disaggregation
In October, CableLabs released new specifications for its Flexible MAC Architecture (FMA) in order to provide vendors with a roadmap to develop interoperable products.
"If you look at the work we're doing with FMA the Flexible MAC architecture, that's one example of what we're doing for disaggregation within the DOCSIS network itself," said CableLabs' Principal Architect Randy Levensalor, who is also on Wednesday's panel. "We kind of started with an integrated CCAP and then we added an RPD (Remote Phy Device) and a virtual CCAP core. That's really disaggregation.
"I'm seeing a lot of the work that ORAN is doing and that 3GPP is doing in the 5G space for defining those new interfaces and new components as well. That really is where you're starting to break up the components. That's putting things onto flexible compute, and that's where we don't limit ourselves to x86 (devices.)
"We're breaking apart additional components with the MAC manager and a lot of the monitoring and all those things are starting to become discrete components."
CableLabs has an open source project called Adrenaline that is working with vendors such as Altran to create a more agnostic pathway for the compute architecture. Levensalor said CableLabs is promoting the idea of using the best compute that's available at the time
"That could be x86. It could be ARM. It could be an FPGA. It could be a GPU. It can just be a smart NIC or some kind of other offloading mechanism," Levensalor said. "So that's really where we're trying to have this all managed largely by some kind of orchestration tool. Kubernetes is what we're seeing as the primary orchestrator for that infrastructure today.
"You can disaggregate with hardware as well as with software, but I think in order to get good virtualization and a broader vendor ecosystem you need that disaggregation because you don't want a bunch of large monolithic applications that don't work well in cloud-native. They don't really take advantage of a lot of the horizontal scaling and failover and all kinds of the nice things that you get with virtualization or cloud-native deployments."
Levensalor said the cable industry has largely been able to skip over the deployment of virtual machines as part of its virtualization efforts by instead going directly to deploying containers.
Lumen tees up disaggregation at the edge
Lumen Technologies is in the midst of rolling out its edge metro nodes that supports a bare metal service ahead of its offering related to x86 compatible device support, according to David Shacochis, vice president and field CTO for enterprise technology at Lumen.
Shacochis, who is also on Wednesday's panel, said Lumen defines disaggregation across two key areas. The first is management disaggregation where the control plane and the data plane are abstracted and separated to enable different scaling factors for each. The second is hardware disaggregation where software is designed to run on general-purpose computing technology and is scaled horizontally.
"Disaggregation builds several bridges," Shacochis said. "Layer 2, Layer3 VPN and OTT/Internet-based connections we see all being used and leveraged by our customers for different application hosting and gateway services. Disaggregation is paving the way for enablement and deployment of private LTE and 5G networks on computing resources within the network, enabling flexible interaction between many different types of operational technology endpoints.
"The key to disaggregation is a flexible enough computing platform that works with various abstraction layers and application containers. Our innovation, architecture and engineering teams are researching and testing disaggregated capabilities within the intent of applying the appropriate results to our platform."
Shacochis said both NFV and disaggregation techniques enhance cloud-services by making NFV capabilities more readily available. In combination with disaggregation, the resources in a enterprise's cloud architecture could be more effectively utilized. Assets across the enterprise network become visible to the service with flexibility to use them when they are needed no matter where they are within the service.
While there are various open source communities that are supporting cloud-native, Kubernetes and disaggregation, there's more to do, according to Shacochis.
"Containerized networking and adoption of containerized approaches still need priority from the vendor community," Shacochis said "We still observe several systems running as virtual-machine guests that could be better scaled leveraging a containerized approach."