---
title: Grooming-Enabled RMSA
url: https://www.emergentmind.com/topics/grooming-enabled-rmsa
type: topic
---

# Grooming-Enabled RMSA

Grooming-enabled Routing, Modulation, and Spectrum Assignment (RMSA) integrates intent-based networking with spectrum resource optimization in elastic optical and IP-optical networks. The key innovation lies in representing network operator connectivity intents and their shared use of underlying infrastructure via a directed acyclic graph (DAG), enabling multiple high-level demands to be consolidated ("groomed") onto the same set of physical and virtual resources. This paradigm shift from disjoint intent trees to intent DAGs enhances spectrum utilization, reduces network blocking, and supports intent-driven automation at the expense of increased architectural and computational complexity [2304.09711].

## 1. Representation and Grooming of Connectivity Intents

Traditional intent compilation separates each network service request into its own hierarchical “intent tree,” where resource allocation is performed independently, precluding resource sharing between demands. Grooming-enabled RMSA replaces this with a unified intent DAG in which all user connectivity intents coexist:

- **Intent trees (no grooming):** Each user intent is represented as a separate root-to-leaf intent tree. Intermediate nodes define progressively lower-level networking actions (lightpath intent, spectrum intent, transponder intent). This structure prevents sharing of resources such as lightpaths or spectrum slots across distinct requests.
- **Intent DAG (grooming):** All user intents are part of a single DAG. Nodes representing shared resources (e.g., a lightpath) may have multiple parent intents, reflecting that several demands are consolidated on a common path. Grooming occurs naturally: if the RMSA solver determines that an existing lightpath is viable for a new demand, the new intent references the shared resource node rather than duplicating it. For example, a LightpathIntent A→C→D→F using slots {5–9} may serve multiple A→F demands at different bit rates.

This DAG structure allows resource sharing, improves allocation efficiency, and underpins the grooming-enabled compilation strategy [2304.09711].

## 2. Mathematical Framework for Groomed RMSA

While the cited work does not provide a complete ILP/MILP formulation, it outlines the primary decision variables and constraints characterizing grooming-enabled RMSA:

| Variable         | Definition                                   | Domain      |
|------------------|----------------------------------------------|-------------|
| $x_p$            | Path $p$ selected (possibly reusing existing lightpaths) | $\{0,1\}$  |
| $y_{p,s}$        | Path $p$ assigned starting at slot $s$       | $\{0,1\}$   |
| $g_{p,e}$        | Path $p$ uses existing lightpath $e$         | $\{0,1\}$   |

Key parameters include demand bit-rate $R$, modulation formats $mod(r)$ characterized by slot requirements $b_r$ and optical reach $d_r$, physical link lengths $L_\ell$, spectrum slots per fiber $S$, and guard band $G$.

Core constraints are:

- **Flow conservation:** Only one end-to-end path is selected per demand.
- **Spectrum contiguity & non-overlap:** Allocated slots must be contiguous for each path and non-overlapping on each fiber, including guard bands.
- **Modulation reach constraint:** Physical path length must be within the modulation format’s optical reach.
- **Grooming consistency:** If grooming onto an existing lightpath, matching slot block and path sequence are required.

These constraints are enforced within the algorithm by the multilayer graph, slot-availability vectors, and intent DAG update rules [2304.09711].

## 3. Adapted Algorithmic Workflow

The approach adapts the Joint Multilayer (JML) and Latency-Driven JML (LDJML) heuristics to operate directly on the intent-DAG structure:

1. **Multilayer graph construction:** A directed graph encapsulates three layers—physical (optical cross-connects and fibers with slot-availability vectors), virtual (routers and installed lightpaths), and inter-layer links (transponders/router ports).
2. **Edge cost calculation:** Each edge is annotated with a cost vector including distance $D_e$, transponder cost $C_e$, router port cost $P_e$, type $T_e$, existing intent reference $I_e$, available modulation formats $\bar{H}_e$, and slot availability $\bar{W}_e$.
3. **Candidate path enumeration:** All feasible multilayer paths $P = \{p_1, ..., p_k\}$ between source and destination are considered, forming valid combinations of optical and virtual traversals.
4. **Pareto filtering:** Non-dominated paths are retained using cost and resource usage criteria.
5. **Path selection:** 
   - JML minimizes combined cost ($C_p + P_p$).
   - LDJML minimizes physical length $L_p$, with cost as a tiebreaker.
6. **Spectrum assignment:** The winning path receives spectrum via first-fit across the available slots.
7. **Intent DAG update:** If grooming onto an existing lightpath is possible ($I_e$ is set), the DAG is updated to reference the corresponding resource; otherwise, new LightpathIntent, SpectrumIntent, and node-level intent nodes are instantiated.
8. **Complexity:** While not exhaustively analyzed, complexity is mainly due to path enumeration, with Pareto filtering and slot assignment being subdominant [2304.09711].

## 4. Resource and Network Model Considerations

The architectural assumptions and resource parameters mirror practical IP-optical network deployments:

- **Spectrum:** Each fiber accommodates $S$ spectrum slots (e.g., $S=368$), with guard band $G$ allocated between adjacent spectrum blocks.
- **Modulation formats:** Each format type is defined by its bit-rate $r_i$, optical reach $d_i$, and required slots $b_i$.
- **Node resources:** Each node supports a finite set of transponders (each incurring cost $C_e$ and reach $D_e$) and router ports (cost $P_e$).
- **Cost vector per edge:** Includes physical distance since last regeneration ($D_e$), transponder/router costs, available modulation formats, slot availability, link type, and existing intent references.

Slot-availability is tracked via bitmaps, and all grooming actions are predicated on exact fiber sequences and slot block alignment [2304.09711].

## 5. Evaluation Methodology

Empirical assessment utilizes the following:

- **Network topology:** Nobel-Germany (17 nodes, 26 fibers) from SNDlib.
- **Traffic demand:** All ordered node-pairs $(i, j)$ receive a random bit-rate drawn from a truncated normal distribution such that the aggregate network load is approximately 62 Tb/s. Each demand generates a unique connectivity intent.
- **Equipment model:** Representative costs drawn from Christou et al. (2022): e.g., 100 Gb/s QPSK (20 slots, 1,000 km reach, 10 k€); 200 Gb/s 16-QAM (40 slots, 500 km reach, 15 k€).
- **Grooming policy:** A new demand grooms onto an existing lightpath if both node sequence and slot block precisely align [2304.09711].

## 6. Quantitative Results and Key Findings

The grooming-enabled RMSA approach yields measurable improvements:

| Metric             | SAP (no grooming)         | JML (cost-driven grooming)      | LDJML (latency-driven grooming) |
|--------------------|--------------------------|---------------------------------|---------------------------------|
| Latency per intent | Lowest                    | Highest (+20% vs. SAP)          | Intermediate (+8% vs. SAP)      |
| Blocking           | ≈12% of intents blocked   | 0% blocked                      | 0% blocked                      |
| Resource cost      | Highest                   | Lowest                          | Intermediate                    |
| Spectrum use       | ~40% slot occupancy       | ~65% slot occupancy             | ~65% slot occupancy             |
| Runtime            | Lowest                    | ~1.5× SAP                       | ~1.5× SAP                       |

Key insights include:

- Grooming via intent DAG eliminates blocking (0% blocking with JML/LDJML vs. ≈12% for SAP).
- Spectrum utilization increases by 25–30% (average slot occupancy from ~40% to ~65%).
- Resource allocation cost is minimized under JML.
- There is a modest increase (~1.5×) in per-intent runtime due to multilayer path search and Pareto filtering, but computation remains in the low-millisecond range.
- Latency can be tuned by switching between cost-driven (JML) and latency-driven (LDJML) heuristics, with LDJML offering a compromise between blocking, cost, and delay [2304.09711].

## 7. Context, Implications, and Significance

The grooming-enabled RMSA approach unifies intent-based service expression with advanced spectrum and path assignment, facilitating both operator intent abstraction and efficient physical resource utilization. By employing DAG representations, the framework supports decentralized, accountable, and confidential multi-domain operation while consolidating traffic to minimize wasted spectrum. However, increased system complexity and a slight increase in per-intent latency are inherent tradeoffs. The technique operationalizes the separation between network operator intent (“what”) and resource allocation mechanics (“how”), furthering the goals of intent-driven networking in practical large-scale optical environments [2304.09711].

Source: https://www.emergentmind.com/topics/grooming-enabled-rmsa