Backbone networks operated by cloud providers serving AI training clusters face a growing disconnect between IP-layer capacity planning and optical-layer physical constraints: IP planners decide how much capacity to turn up, but without visibility into spectrum fragmentation, channel placement, ROADM connectivity, and per-band heterogeneity introduced by multiband (C+L) transmission, they produce plans that require additional iterations to converge at the optical layer. Solving the joint IP-optical optimization is computationally prohibitive at production scale. We present FABRIC (Fragmentation-Aware Backbone Resource and IP Capacity planner), a cross-layer spectrum planning system that models spectrum at the lowest frequency-slot granularity, groups IP links by shared-risk link group, computes fragmentation losses from channel data, and feeds these constraints into a weekly capacity optimization cycle. We evaluate FABRIC and show that spectrum-aware planning accurately produces capacity plans grounded in optical-layer reality.

Network design is the process of dimensioning IP capacity over an optical network infrastructure to satisfy a given set of demands and reliability constraints. Specifically, we consider the problem of hose-based cross-layer network design, which seeks to find a minimum cost design that is able to route demand for all hose traffic matrices under all specified failure states. While most network design problems are solved as Mixed Integer Programs, a commercial solver can become intractable due to the scale of today's networks. We demonstrate how the classic Benders decomposition algorithm can be applied and improved for this problem and discuss practical implementation aspects. We showcase a horizontally scalable distributed framework to leverage the decomposable problem structure and solve millions of linear programs in a distributed manner, thereby making the network design problem tractable. In contrast to the conventional approach where failure states and traffic matrices are planned sequentially, the Benders algorithm finds global optimal designs across all traffic matrices and failure states. This leads to network designs with improved solution quality and reliability, with 20--30% less IP capacity and spectrum consumption, 50% less link augments and up to 20x faster runtime that enables design for hyper scale networks in a matter of hours.

Traffic grooming continues to be a rich area of research in WDM optical networks. In this paper, we identify four kinds of path level traffic aggregation/grooming strategies - point to point (P2P), point to multi-point (P2MP), multi-point to point (MP2P), and multi-point to multi-point (MP2MP). We develop an auxiliary graph based approach as a generic network dimensioning solution for these architectures. Using our model, we compare the performance of various architectures employing path level aggregation in both single hop and multi-hop scenarios. For the studied scenarios, we observe the following - (1) MP2MP outperforms other architectures by multiple orders of magnitude in single hop networks, (2) P2P performs the best in multi-hop transceiver constrained scenarios, and (3) P2MP performs the best in multi-hop wavelength constrained scenarios.

Recent research has focused on designing new architectures and protocols for pure optical grooming without resorting to fast optical switching. This is achieved in three steps: (1) the circuit is configured in the form of a path or a tree (2) optical devices like couplers/splitters are used to allow a wavelength to aggregate traffic from multiple transmitters and/or receivers through one of the following means - point to point, point to multi-point, multi-point to point and multipoint to multi-point (3) an arbitration mechanism is provided to avoid contention among end users of the circuit. In this paper, we focus on design of mesh networks that aggregate traffic at the path level. We develop a shared, mixed protection algorithms for guaranteed survival from single link failures in the context of dynamic traffic for mesh networks. Based on our simulations on random graphs, we conclude that light-trails, that enable multi-point to multi-point aggregation, performs multiple orders of magnitude better than lightpaths that allow point to point aggregation.

Traffic grooming continues to be a rich area of research in the context of WDM optical networks. We provide an overview of the optical and electronic grooming techniques available with focus on IP as the client layer. We discuss the various architectural alternatives available: peer, overlay and augmented models. We first provide a survey on the research work in the area of traffic grooming in optical circuit switched networks. We then identify problems with electronic grooming in terms of high speed router design and bring out the merits of optical grooming. Next, we describe the shared wavelength optical network technology called light-trails and compare its performance with electronic grooming networks for both the peer and overlay models. Based on our simulations on random graphs of various diameters, we identify the threshold router speeds at which light-trails can compete with the electronic grooming solution for a given network scenario. We conclude that since the present router capacities are below the threshold speed or such routers are likely to remain expensive for some time, light-trails is an appealing candidate solution.

We investigate the characteristics and performance of architectures that allow statistical sharing of wavelengths in the optical layer. The bandwidth of a light-wave circuit can be shared in the optical layer by multiple nodes that are along a linear path in a network through the inclusion of simple but flexible hardware overlaid with control software. We consider three types of architectures - lightpath (LP), light-trail (LT) and source based light-trail (SLT) networks. The three differ in their level of optical sharing, in their hardware and software requirements and in their performance. We study the interplay between statistical sharing at the optical layer and circuit switched sharing at the electronic layer. We develop a generic auxiliary graph model for traffic grooming in heterogeneous WDM mesh networks that can accommodate constraints related to transceivers, wavelengths, and groomers for architectures with different levels of statistical sharing and for both static and dynamic scenarios. We conclude based on our simulation results that sharing wavelength in the optical domain is beneficial and can lead to reduced blocking

We explore network design principles for next-generation all-optical wide-area networks, employing light-trail technology. Light-trail is a light-wave circuit that allows multiple nodes to share the optical bandwidth through the inclusion of simple but flexible hardware overlaid with a lightweight control protocol. We develop light-trails as a novel and amenable control and management solution to address IP-centric communication problems at the optical layer. We propose optical switch architectures that allow seamless integration of lightpath and light-trail networks, and assess their costs and capabilities. We formulate the static light-trail RWA problem as an integer linear program. Since this programming problem is computationally intractable, we split it into two subproblems: (a) trail routing, for which we provide three heuristics, (b) wavelength assignment, for which we use the largest first heuristic available in literature. The objective of our design is to minimize the optical layer and electronic layer costs in terms of the number of wavelengths and communication equipment required. We illustrate our approach by comparing the performance of our trail design heuristics on some test networks.

Recently, a new architecture called light-trails has been proposed that provides a novel control and management solution to address IP-centric issues at the optical layer. By inclusion of simple hardware that performs drop and continue functionality, overlaid with a light-weight control protocol, light-trails enable efficient sharing of network resources, improve bandwidth utilization and minimize network costs. Due to power budget constraints in such networks, it may not always be possible to have end to end communication in pure optical domain and requests may be required to traverse multiple intermediate transit points called hub nodes before reaching the final destination. The hub nodes need to be equipped with special hardware for switching and grooming connections. We investigate the problem of designing networks where such hubs are sparsely located. We show through our simulation results that by carefully designing heuristics for hub node placement and trail routing, it is possible to achieve high throughput with minimal number of hub nodes.

The light-trail architecture provides a novel solution to address IP-centric issues at the optical layer. By incorporating drop and continue functionality, overlaid with a lightweight control protocol, light-trails enable efficient sharing of network resources, support subwavelength traffic and minimize costs. In this work, we investigate network design and survivability issues in such networks in the presence of multi-granularity subwavelength traffic subject to nonbifurcation constraints. We first establish the NP-Hardness of the light-trail routing problem by reduction from a Hamiltonian path problem. We propose three heuristics for lighttrail network design and study their performance with limited network resources. We observe the effect of tunable and fixed transceiver equipment on network throughput. We observe that our heuristics yield excellent wavelength utilization under moderate to high loads even in the presence of heavily fractional traffic.We propose two additional heuristics for shared and dedicated protection and conclude that with only a modest amount of spare capacity, full protection can be achieved for all single link failures.