June 17, 2020

Chen Goldenberg

Head of Systems Engineering

What’s Driving Service Providers to Network Disaggregation?

Declining revenues and lower margins have put network service providers profitability at risk. At the same time, demand for capital expenditure has been increasing with the need to invest in both existing infrastructure to meet the rapid growth in data traffic, and also to find new areas for growth.  The pressure to substantially grow network scale, while controlling its cost, has created an inflection point for service provider networking, especially as seen under the current pandemic. Once, I was at the airport during a network outage and everything was brought to a standstill. There were no landings and no take-offs. It made me think, how could we prevent this from happening?

What’s Driving Service Providers to Network Disaggregation?

Presenting our solution and our distributed NOS in detail at Networking Field Day 22, I explained why service providers are looking to take the disaggregation approach to address the needs of todays high scale networking. While high-scale disaggregated routers suitable for core and edge networks offer a new foundation for the market to build an open and more profitable model, from a technological perspective it still included challenges.

Options for Building a High-scale Networking Solution

In considering different approaches to invest their resources, service providers had several different paths to consider:

  • The chassis-based platform – running networks for the last 30 years – had merits, but also disadvantages.
  • CLOS architecture with the spine and leaf structure – primarily used by webscalers in data centers.
  • Disaggregated networking – a new distributed disaggregated model that separates software and hardware.

Service Providers Criteria for a Networking Solution

We took a look at service providers across the globe, from Tier 1 to Tier 3, and set down their most important criteria for considering a solution that will address their network challenges.

This is what they said:

  • Reliability. The network must be available all the time, no matter what happens. The solution must not have any single point of failure, it must be able to recover from even multiple simultaneous failures, and the software must be smart enough to give you a heads up before an issue occurs, so you can prevent it from happening altogether.
  • Scale and capacity. SPs plan their networks’ growth 5-10 years in advance and they want to avoid making expensive changes in the networks’ architecture along the way. But how can they plan ahead, when they don’t know what the demand for capacity will be? SPs don’t want to just add large hardware investments as the need arises.
  • Simplified. With dozens of boxes at play, SPs need an automated management process (a zero touch network) so that fewer people are required to ensure network operations.
  • Cost-effective. Not only does the cost of the hardware need to be lower, but also the OPEX to run it. The solution must meet the business goals.

How do Different Network Approaches Measure Up?

To maintain or improve profit margins, service providers needed an approach to IP networking that allows them to cost-effectively scale to meet growing traffic requirements. The options before them were taking a monothlithic chassis approach, a CLOS model, or the disaggregated model.

  • Monolithic chassis
    Although this is a well known ruggedized design that has been field proven since the networks’ early days, it nevertheless requires extensive planning to accommodate the unknown capacity that will impact the network. There are far less chassis sizes compared to site sizes. Therefore, buying hardware in advance is expensive. If SPs under-provision, they will need to literally forklift in equipment to upgrade. If, on the other hand, they over-provision, they will be underutilizing the hardware, while having an expensive outlay upfront. And, of course, once the chassis is purchased, parts cannot be mixed and matched from different vendors, compounding expenses and creating vendor lock.
  • CLOS architecture
    With the spine and leaf structure it is easy to scale out and up and grow almost infinitely. With this solution, you can mix hardware from different vendors as the connectivity between the boxes is over standard protocols. The CLOS model is perceived as more redundant because each box is a standalone router and with so many control planes running separately, any failure can be overcome through the convergence of the routing protocol.However, this model relies on large bundles, which are notoriously inefficient, as the layer 3 hashing is unable to balance the load and hot-spots begin to appear where some bundle members are over-utilized and some are under-utilized. Additionally, in the CLOS model, the propagation of the IGP routes and link states is known to only scale to the limits of the IGP capabilities. Moreso, each box in this model is a standalone router that is configured and managed separately, which does not allow for an end-to-end view, so you can easily run into congestion.

The Best of Both Worlds: Network Disaggregation

Considering SP’s requirements, adapting the disaggregation model for networking is very attractive. The disaggregated model brings together the advantages of the chassis and CLOS models, while solving their pain points.

Disaggregated Network ModelDriveNets Network Cloud is based on a distributed disaggregated router architecture that can reach extreme capacity by scaling linearly from a single white box router of 4 Tb/s to a cluster of hundreds of white boxes (up to 768 Tb/s) – acting as a single router entity. This scalability of a single router entity over any number of white boxes allows it to support any location in the network – Core, Edge, and Access with a single software based on only two white box building blocks. This model allows the mix-and-match of the hardware in order to bring prices down as you scale-out. The white boxes used in the cluster can be spread across multiple racks to accommodate your unique cooling and power capabilities. Also, with the chip’s virtual output queues capabilities, which provides end-to-end visibility, you can distribute the services (e.g. broadband, business, critical services, etc.) within the cluster, even though the cluster acts as a single routing entity. The chip’s forward and backward compatibility allows you to introduce newer chip derivatives into the cluster with the knowledge that they will work with the existing cluster components.

Market Adoption of the Disaggregated Router?

What is the state of the disaggregated router in the market today. With the ability to scale in a cost-effective way, DriveNets is already working with Tier 1 operators on two use cases: core router and the edge. We are also developing additional use cases with prospective customers, including an aggregation router, mobile backhaul for 5G, and a cell site router.

There is really good traction with SPs, having taken a chance on a relative newcomer like DriveNets. They have been impressed with our disaggregated model, seeing how it reduces the total cost of ownership (TCO), scales to almost any size, offers power efficiency, and allows you to only pay for the required hardware as needed. We enable service providers to support current and future network capacity demands with a simple, cost-effective and telco-grade open disaggregated network architecture, with the lowest TCO.

Presented at the Networking Field Day, February 12, 2020

AvidThink White Paper

Network, Meet Cloud - Scaling Disaggregated Network Infrastructure using Cloud Principles