Managing the growing cost of optics in networks
In this edition of Voices of the Industry, Bob Shine, Vice President of Marketing and Product Management at Telescent, explores how the increasing costs of optics in networks can be managed through dynamic allocation of resources at the level of the fiber layer.
As optical networks have evolved to higher data rates to handle astronomical increases in traffic, the percentage of network costs associated with optical components has increased dramatically. Finding ways to economically manage this growing percentage of optical costs has become a priority for many network operators and providers. Dynamic allocation of resources at the fiber layer is one way to improve the economics of the optical network and can be implemented today using large scale robotic fiber interconnects.
The growing share of optical component costs in networks is driven by the need for networks to support higher port densities and faster speeds. This results in the need to use higher speed optics as shown by the transition from 40G to 100G and now to 400G. Moore’s Law reduced the cost / bit for switches and routers through advanced silicon ASIC development. However, while the cost / bit for pluggable optics has also fallen, it has not fallen nearly as quickly as the costs of the corresponding silicon components.
Based on the economic trends above, Cisco said that for 10G networks, optics is about 10% of the total hardware cost of a data center network. On the other hand, optics will represent more than 50% of the total cost of the material for a 400G network. [Optics in the Data Center: Powering Ever-Increasing Capacity Demands – Cisco Blogs]
As you can imagine, this cost challenge occurs in a wide variety of areas – essentially any situation where compute and storage components are associated with high speed transmission requirements. These include long distance and now 5G networks, data center networks, and even machine learning resources within data centers. Backbone optical networks were often the first areas of the network to migrate to higher data rates due to the sheer amount of traffic transmitted over limited and expensive long distance fibers. The densification of 5G networks and front-haul requirements towards radio towers increase the cost of the optical network for 5G. In data centers, moving with growth while minimizing capital costs requires efficient ways to add capacity that can be managed through dynamic fiber management. Even applications such as machine learning, which is very computationally intensive, will see an increasing share of optical component costs due to the need to transfer ever-increasing sets of parameters between learning iterations. The following sections will show how all of these cases can benefit from dynamic resource allocation at the optical layer.
The advantage of dynamic fiber cross-connections for long distance networks
For basic optical networks, a group of AT&T showed significant cost savings and improved robustness through the use of software-defined network control (SDN) and dynamic fiber cross-connections (DFCC) to provide joint optimization of the IP and optical layer. Essentially, SDN and DFCC allow disaggregation of a failed link, allowing functional components of the failed link to be reused in a redesigned route. For a central office with two large main routers, rather than taking both a router port and a transponder offline if a component fails or requires a software restart, a dynamic fiber cross-connection allows the working component to go offline. be recombined with a port in the second router. Even without failure, the DFCC can be used to recombine router / transponder pairs to create more efficient routing. Adding the only automatic fiber optic connection in this use case resulted in an estimated 5% savings across the entire network stack.
The advantage of dynamic fiber cross-connections while increasing data center capacity
Moving Inside a Data Center, while the construction of a data center must be completed before commissioning, data centers are usually slowly filled with servers and switches. This allows the capacity to meet growing demand while avoiding idle capacity. The initial medium-sized network will need to be extended while the network carries over-the-air traffic. While other network topologies have been investigated to allow for fine-grained expansion, it is possible to extend using a traditional Clos network using a new topology-based methodology, as demonstrated by work done at Google. Live expansion requires maintaining sufficient network throughput for the duration of live expansion to avoid congestion. This is done through several automated steps where a limited subset of network elements are disconnected and added. Since the capacity is removed during these rewires, each step should be completed as quickly as possible. To simplify reconfiguration, all server blocks and backbone blocks are connected through a group of patch panels – creating a DCN topology that can be created and changed simply by moving the fiber jumpers on the patch panel. Managing this rewiring during upgrades can be greatly simplified and managed with a robotic fiber cross connect system.
The advantage of dynamic fiber cross-connections for machine learning resources
Finally, machine learning inside data centers is an increasingly important part of compute capacity. While the use case is dominated by the development of advanced GPU and TPU processing capabilities, the very large and growing training sets used create a high demand for communication bandwidth between GPUs. NVIDIA’s fastest ML training platforms deliver 1.2 Tbps of communication bandwidth between its GPUs. But while conventional data center workloads behave unpredictably with short streams dominating the workload, machine learning workloads are predictable, sparse, and mostly consist of large transfers. By taking advantage of these features, a dynamic fiber cross-connection that adjusts the communication bandwidth between GPUs based on the training model can dramatically improve performance.
An example of a dynamic fiber crossover to meet the above use cases would be a Network Topology Manager (NTM). When selecting an NTM, look for one that uses a robot to remotely configure and reconfigure the interconnects within minutes. Ideally, you’ll also want to select a pay-as-you-grow NTM model that can scale the number of ports while preserving the connectivity capability of everything to everything. You’ll also need a system with low-loss, locked-down connections, and a fully maintainable system in the field without disrupting traffic. Finally, your NTM must be NEBS Level 3 certified to guarantee the reliability demanded by customers.
It is certain that data traffic will continue to grow, which will increase the need for increased bandwidth at all levels of the data network. While cost containment will require progress in a number of areas, adding dynamic resource management at the fiber layer is a way to more effectively cope with the growing share of optical costs in the network. .
This article was written by Bob Shine, vice president of marketing and product management, Telescent. Telescent’s Network Topology Manager (NTM) uses a robot to remotely configure and reconfigure cross-connections in minutes. It provides dynamic control at the fiber layer, is NEBS certified and available to meet your current needs. For more information, contact Telescent today.